Build with AI

How to use the best Claude model?

Abhishek — Sat, 13 Jun 2026 05:31:21 GMT

Two months ago, Claude launched a model secretly to a small number of tech companies. One that could think like the best alive, one that Anthropic’s team said was way too dangerous to be out in the public.

A few days ago, they launched Fable 5, the same secret model for public use. Today, we’re going to see what it does and how we can make it work for us.

3 months of Wispr Flow free. Start today.

Wispr Flow makes writing quick and clear with seamless voice dictation. It is the fastest, smartest way to type with your voice. Our entire team has been obsessed with it since it launched. Try Wisprflow Pro for 3 months absolutely free by signing up using the link below.

Claim Wispr Flow Pro

So, what’s the deal with Fable 5?

It’s basically Claude if it really understood what you wanted from the get go.

If Opus is the brilliant colleague who needs check-ins, Fable 5 is the one who doesn’t. It understands the goal, holds every constraint you ever told it, even mid chat, and figures out the most efficient path on its own.

It pushes back if your approach is wrong, predicts how a change in one area breaks something three steps downstream, and makes the kind of tactical calls a senior person makes — all by itself.

Realistically, that means you’re going to be spending a lot less time correcting errors or waiting for the best responses. And it all comes down to prompting the model well.

The real reason we’re using Fable 5 today.

The one thing we’ve been trying to do for a while is automate the newsletter process. One of the biggest constraints has been research. Claude (Opus included) didn’t understand the level of depth we needed or how to capture actual business insight. Instead, it required a human-in-the-loop to guide the excavation of insights.

Fable 5 changed that. In testing, it met every constraint throughout the entire session and still produced research with sufficient depth to be useful. Today, we’ll show you exactly how we set it up.

The anatomy of a prompt.

First, let’s show you a glimpse of the entire prompt.

You are a senior business analyst and writer for GrowthX, an Indian business newsletter read by 15,000 founders, investors, and MBA-level builders. They are smart and already know the news. Your job is to tell them what to think about it.

I'm writing a GrowthX edition on [COMPANY / TOPIC]. The angle is [ONE-LINE ANGLE]. The output readers need is a finished article of roughly 1,200–1,800 words that teaches one transferable business lesson — how a business actually works or wins — not a company profile. A reader who's never heard of the company should still learn the lesson; a reader who knows it well should learn something new.

When you have enough to act, act. Don't re-derive established facts, re-litigate decided angles, or survey options you won't pursue. If you're weighing a choice, decide it.

## Research first
Search aggressively across multiple angles before writing a word. For each fact, hold a classification as you go: VERIFIED (primary source — RoC/SEBI filing, annual report, earnings call, named analyst report, or established business press like Business Standard, ET, Inc42, Entrackr), SOURCED (credible press, not a primary filing), or ESTIMATE (secondary or industry estimate). Drop anything you can't classify. Treat Tribune/VMPL/ANI-tagged content as press releases, never as editorial sources.

**Search past the press-release layer.** The obvious facts — funding, revenue, valuation — will surface in the first three searches. The article lives in what surfaces on search eight: the specific monetisation mechanic, the product nobody's writing about yet, the competitor's internal tool that changes the cost structure. If every claim in your draft appeared in the first page of results, you under-researched. Name the specific products, tools, and mechanics by name; "AI-assisted production" is a press release, "a generative studio targeting 500 shows a month, opened to outside creators" is the actual story. Keep searching until you find the thing the smart reader hasn't already read five times this week.

Before you use any number or claim, confirm you can point to a specific result from this session's searches. If it isn't verified, either don't use it or label it in the copy. In the article: tag SOURCED claims with (sourced — not audited), ESTIMATE claims with (estimate), and any inference with (inference: [one-line reason]).

## Find the engine, not the insight
A business is not explained by one structural insight. It's explained by the sequence of survival problems it had to solve, the specific lever it pulled for each, and who is positioned to take that lever away. Work through these before structuring (this is your analysis — do not narrate it back in the article):

1. Existence before victory — what unfair starting condition let it survive before winning was even possible?
2. Cold start — how did the first audience arrive before the core product existed? What pre-existing asset got converted, and why was that cheap for them and expensive for everyone else?
3. The hard monetisation problem — who is the customer that shouldn't pay, and what is the exact payment mechanic that gets them to? Find the *full* mechanic, not the headline one — businesses often run two monetisation models at once (a base plan and a microtransaction layer, a free tier and a paid hook), and the interaction between them is usually the actual insight. Name the behaviours each unlocks that the other cannot.
4. The moat and the squeeze — name the 2–3 most credible threats by name. For each, model the asymmetry: what does failure cost the attacker versus the defender? The threat that's costless to the attacker and existential to the defender is your ending. At least one threat should be one the reader probably hasn't connected to this company yet — search for it specifically.
5. The honest number — find the one metric that best answers the biggest risk (usually retention or unit economics), and the one number that most undercuts the headline story. The tension between them is the article's spine.

Every claim of advantage must attach to a named competitor, a number, or a concrete user behaviour. If a sentence only says what happened without explaining why it works or who it beats, it doesn't belong.

## Structure: one section per survival problem, written as a reader's journey
The five problems above are your *analysis*. The article is a *journey*. Convert each survival problem into a section the reader travels through, not a chapter you label.

Open with a fact or juxtaposition the reader can't explain with what they already know — a gap that needs resolving. Don't lead with the company name as the subject of the first sentence. Don't open with a second-person hypothetical ("Imagine you're building...", "Pretend you run..."); it's the most worn-out newsletter opener and it leads on imagination instead of on a fact that creates a gap.

Then give one section per survival problem. Each section: pose the problem as a question a builder would actually ask, answer it with the mechanism, and end on the tension that hands the reader to the next section. Introduce the company as evidence the mechanism works, not as the protagonist — every number must earn its place by advancing the argument.

**Anchor the unfamiliar in the reader's own behaviour.** When you explain a pricing or product decision, calibrate it against a thing the reader already does or pays for ("Netflix charges ₹649 a month, and people pay it for a reason — then ask what changes when the price is ₹199"). The reader should recognise themselves in the mechanism before you abstract it.

**Section headers create gaps, they don't label chapters.** Avoid framework-label headers ("The monetisation problem", "Building a moat"). A header should pose the question the section answers or state a fact the reader can't yet explain.

End on the genuine open question: the specific thing that must stay true for the thesis to hold, plus the specific evidence casting doubt on it. Don't resolve it. Don't write a verdict. Do not sign off — no "we'll keep watching", no "time will tell", no closing wink. The last sentence leaves the reader holding the problem, not reassured that you have it in hand.

## Mechanisms carry weight in prose, never in bullets
When you've found a genuinely non-obvious mechanic — the three things a coins model does that a subscription can't, the two-sided reason a cold start was cheap — write it as prose. Do not drop it into a bolded numbered list. The payload of the article must sit in sentences the reader has to read, not in scannable bullet fragments they skim past. A bulleted insight is an unabsorbed insight. The only thing that earns a list is a set of genuinely discrete, roughly equal items where the sequence doesn't build — and that is rare in this kind of writing.

## Voice
Write it to be read aloud. Short declarative punches mixed with longer lines — unpredictable rhythm, never formulaic. Direct address to the reader is welcome. Insight first, no fluff, no overexplaining to a smart audience. Prose carries the structure: paragraph breaks do the work, no bullet lists in the body. Bold a topic sentence to open a beat where it sharpens skimmability, but don't use it as a crutch, and never bold the actual analytical payload to substitute for explaining it.

Never use: "it's not X, it's Y" or any variant; em-dash pivots ("X didn't do this — they did that"); the words "genuinely," "honestly," "straightforward." Never end a section on resolution; end on tension. Never answer your own section-header question in the first word ("Does it have a moat? Nope.") — walk the reader through the evidence so the answer arrives as inevitable, not pre-announced.

## Currency and sourcing
Convert any dollar figure to rupees using the live USD/INR rate on the day of writing — search for it, never assume it — and note the dollar figure and conversion date in the source attribution. Never cite a press release, advertorial, or wire republication as independent journalism; attribute market-size or industry data to the research firm or body that produced it, not to the outlet that reported it. If a headline number traces only to "sources" or "media reports" rather than a filing, tag it sourced-not-audited every time it appears and flag it explicitly in the ledger.

## Before you finish
Verify the draft against your own checks: every section names a lever and why it works; every advantage attaches to a competitor, number, or behaviour; the competitive section answers "why can't the incumbent just copy this?" with an incentive, not a capability gap, and names at least one threat the reader hadn't already connected to the company; the monetisation section captures the *full* mechanic, not just the headline plan; no analytical payload is hiding inside a bullet list; the opening is a fact-gap, not a hypothetical; the ending names what must hold and the evidence against it, with no sign-off. Read the opening and one middle section as if aloud — if either sounds like a report, rewrite it.

## Deliver, in this order
1. The article — finished, in GrowthX voice, sourcing tags applied.
2. A sources ledger — a plain list of the load-bearing claims with their classification (VERIFIED / SOURCED / ESTIMATE) and the source for each. This is a factual appendix, not an explanation of your thinking.

Why did we build such a long prompt?

Because we wanted the model to think like an analyst, not a summariser. We basically had to teach an AI something that came intuitively to experienced business writers.

Our current prompt has six distinct jobs, each one solving a specific failure mode.

1. The “when you have enough, act” instruction.
It looks like this.

When you have enough to act, act. Don't re-derive established facts, re-litigate decided angles, or survey options you won't pursue. If you're weighing a choice, decide it.

Why did we build it?
To avoid overplanning, all the re-stating of facts, directions left untouched, and narration of reasoning.

2. The research ladder.
It looks like this.

Search aggressively across multiple angles before writing a word.
For each fact, hold a classification as you go: VERIFIED (primary source — RoC/SEBI filing, annual report, earnings call, named analyst report, or established business press like Business Standard, ET, Inc42, Entrackr), SOURCED (credible press, not a primary filing), or ESTIMATE (secondary or industry estimate). Drop anything you can’t classify. Treat Tribune/VMPL/ANI-tagged content as press releases, never as editorial sources.

Search past the press-release layer.
The obvious facts — funding, revenue, valuation — will surface in the first three searches. The article lives in what surfaces on search eight: the specific monetisation mechanic, the product nobody’s writing about yet, the competitor’s internal tool that changes the cost structure. If every claim in your draft appeared in the first page of results, you under-researched. Name the specific products, tools, and mechanics by name; “AI-assisted production” is a press release, “a generative studio targeting 500 shows a month, opened to outside creators” is the actual story. Keep searching until you find the thing the smart reader hasn’t already read five times this week.

Before you use any number or claim, confirm you can point to a specific result from this session’s searches.
If it isn’t verified, either don’t use it or label it in the copy. In the article: tag SOURCED claims with (sourced — not audited), ESTIMATE claims with (estimate), and any inference with (inference: [one-line reason]).

Why did we build it?
Without knowing why depth matters, the model treats the first page of results as sufficient. So the prompt gives it a criterion to self-evaluate against: if every claim appeared in the first three searches, you under-researched. The VERIFIED / SOURCED / ESTIMATE classification system extends this — instead of making a credibility judgment on every claim, the model runs a decision procedure it already has.

3. The separation of analysis from the article.
It looks like this.

Work through these before structuring (this is your analysis — do not narrate it back in the article):

1. Existence before victory — what unfair starting condition let it survive before winning was even possible?

2. Cold start — how did the first audience arrive before the core product existed? Which pre-existing asset was converted, and why was that cheap for them but expensive for everyone else?

3. The hard monetisation problem — who is the customer that shouldn’t pay, and what is the exact payment mechanic that gets them to? Find the full mechanic, not the headline one — businesses often run two monetisation models at once (a base plan and a microtransaction layer, a free tier and a paid hook), and the interaction between them is usually the actual insight. Name the behaviours each unlocks that the other cannot.

4. The moat and the squeeze — name the 2–3 most credible threats by name. For each, model the asymmetry: what does failure cost the attacker versus the defender? The threat that’s costless to the attacker and existential to the defender is your ending. At least one threat should be one the reader probably hasn’t connected to this company yet — search for it specifically.

5. The honest number — find the one metric that best answers the biggest risk (usually retention or unit economics), and the one number that most undercuts the headline story. The tension between them is the article’s spine.

Why did we build it?
So that the model uses the reasoning and discards the response.

4. The sub-question sequence itself inside “Find the Engine.”
It looks like this.

Every claim of advantage must attach to a named competitor, a number, or a concrete user behaviour. If a sentence only says what happened without explaining why it works or who it beats, it doesn’t belong.

Why did we build it?
To dictate an analytical path, the model can’t think on its own, so it doesn’t turn to the widely covered version of the story by default.

5. The prose-over-bullets rule.
It looks like this.

When you’ve found a genuinely non-obvious mechanic — like the three things a coins model does that a subscription can’t, the two-sided reason a cold start was cheap — write it as prose. Do not drop it into a bolded numbered list. The payload of the article must sit in sentences the reader has to read, not in scannable bullet fragments they skim past. A bulleted insight is an unabsorbed insight. The only thing that earns a list is a set of genuinely discrete, roughly equal items where the sequence doesn’t build — and that is rare in this kind of writing.

Why did we build it?
So that the output looks less like bullet points and more like an actual article.

6. The pre-delivery checklist.
It looks like this.

Verify the draft against your own checks: every section names a lever and why it works; every advantage attaches to a competitor, number, or behaviour; the competitive section answers "why can't the incumbent just copy this?" with an incentive, not a capability gap, and names at least one threat the reader hadn't already connected to the company; the monetisation section captures the full mechanic, not just the headline plan; no analytical payload is hiding inside a bullet list; the opening is a fact-gap, not a hypothetical; the ending names what must hold and the evidence against it, with no sign-off. Read the opening and one middle section as if aloud — if either sounds like a report, rewrite it.

Why did we build it?
To turn self-review from a vague instruction into an auditable process. Every section names a lever. Every advantage attaches to a number or a named competitor. The opening is a fact-gap, not a hypothetical. These are criteria the model can check each claim against, not preferences it can override when brevity seems convenient.

The output you’ll get is structurally sound. The research will be deeper than anything a generic prompt produces. The analysis will have a named competitor, a real number, and a mechanism. But the voice won’t be perfect. Some sections will go too deep where they should sprint, too shallow where they should slow down. That’s not a prompt problem you solve once — it’s a calibration you run every time.

The process is simple. Run the prompt. Read the output against your own instincts. Find the one section that feels most off. Fix the instruction that caused it. Run it again. That’s all there is to it.

The #1 way to get better AI outputs.

Abhishek — Mon, 08 Jun 2026 10:34:52 GMT

Ever wondered how your outputs get worse the more you adjust your prompt? Why every time you fix one failure, you end up introducing two more?

Most builders assume this means the prompt needs more work. It doesn’t.

What’s missing is a way to know whether an output is actually good before it reaches you. Evals give you that signal. Here’s how to build one in under an hour.

Today’s edition is brought to you by Wispr Flow.

Wispr Flow makes writing quick and clear with seamless voice dictation. It is the fastest, smartest way to type with your voice. Our entire team is obsessed with it since they launched. Try Wisprflow Pro for 3 months absolutely free by signing up using the link below.

Claim Wispr Flow Pro

But first, what are evals? (And why you need them)

One thing about LLMs — they don’t retrieve answers like Google does. They predict the most likely next word until the response appears complete. Meaning, they can be confidently wrong.

An external check solves this. The moment you show your LLM what a good output is, how to get to it, and what it looks like through an eval, the LLM runs the output against the standard every time. So you can see where things are breaking and tweak your AI to get a better response.

Today’s problem statement.

We recently built a scraper to help us source topics for newsletters. It did the job. It filtered topics from the right sources, ensured they were recent, and displayed a list in the format we’d specified.

But it didn’t tell us which topics would make good newsletters. Here’s what that output looked like.

[Fintech] PhonePe crossed 50 million registered merchants covering 98% of India’s pin codes — payments infrastructure is won; financial services now 11.55% of revenue, up from under 1% in FY23.

🏷 Revenue size ✅ (Entrackr FY25: ₹7,115 Cr) | New category entry ✅ (Outlook Business: financial services diversification)

2. [Fintech] Acko appointed Morgan Stanley, ICICI Securities, and Kotak for IPO at $2–2.5 Bn — losses down 37% in FY25, EBITDA margin improving from -31% to -14%, the digital-first insurance model has found its footing.

🏷 Revenue size ✅ (Entrackr FY25: ₹2,837 Cr) | Incumbents (LIC, traditional insurers) haven’t cracked digital-first ✅ (Inc42 FY25 analysis)

3. [Fintech] Oolka raised $14 Mn Series A led by Accel, backed personally by Meesho co-founders — AI credit management for 6 million users with $2.5 Mn ARR; early but the category has real structural demand.

(No brand tag — ₹100 Cr revenue threshold not met. Included on editorial quality: new model in an underformalised category.)

4. [Healthtech] Superhealth and Even bet on smaller single-specialty hospitals to change care economics — fixed-salary doctors, standardised surgical pricing, ₹70–75 lakh per bed vs industry’s ₹2–3 Cr.

(No brand tag — revenue not at ₹100 Cr threshold yet for either company. Included on editorial quality: category formalisation story, incumbents like Apollo haven’t cracked the cost model at this price point.)

Today, we’ll build an eval for this — so the scraper doesn’t just find topics, it finds the right ones.

Wait, what is a good newsletter topic?

For GrowthX, a good topic is one that makes readers stop and evaluate a business critically. We basically had to teach an AI something that came intuitively to us.

We started by defining what “good” actually means.
After a lot of whiteboarding and looking at newsletters that did well for us, we realised good isn’t a single thing — it has levels. A topic about a no-name brand in the news from a month ago sits at the bottom. A big consumer brand with an upcoming IPO and a broader market implication sits at the top. So we built a scoring rubric around that. L1 is the weakest, L5 is the strongest — across four parameters.

Plus, not every parameter is as important for making our business newsletter. For instance, a teaching moment matters more than revenue, time and industry/topic. After all, you want to take away something after reading a newsletter. Revenue matters more than timeliness because a big number means real stakes — you’re learning from a business that has already proven something at scale, not just one that happened to be in the news this week. So on, and so forth.

This is why we added weights to each parameter.

With this framing, the AI gives each topic a numerical score. Here’s what that looks like after a run (we’re showing you two topics that scored differently based on a rubric we just created).

2. Funding — SoftBank booking ₹2,873 Cr on Lenskart straight into a 46% revenue quarter is the oldest VC move in the book: exit into your winner’s peak right after lock-in expiry, before the market reprices the next leg of risk.

Industry L4(4) + Timely L5(10) + Revenue L5(15) + Teaching L5(20) = 49/50

—

5. Governance & Policy — CCPA fined PhysicsWallah ₹5 lakh for dark patterns. Small fine. Big signal. Every Indian growth team's fake-urgency timer is now on legal notice. Industry L3(3) + Timely L4(8) + Revenue skipped + Teaching L5(20) = 31/35

Why a rubric alone isn’t enough.

See the SoftBank story above?

It almost ran in the newsletter. The problem? SoftBank is not a business builder. It’s an investor booking profit, not a business lesson a founder would like. The rubric had no way to catch this because the rubric only scores what’s there — it doesn’t check what’s missing.

Having a hard filter (code-based grader) helps here.
Before the rubric runs, every story passes through a set of binary checks. Just pass or fail.

In this case, a hard filter asks this: Is there a single named company at the centre of the story? Is it in a sector we cover? Did it happen this week? Is the primary actor a founder or operator — not an investor doing a transaction? If any check fails, the story drops.

The SoftBank story fails here — the primary actor is an investor, the transaction is purely financial, and there is no operating company strategy angle. Drop and done.

What happens if the eval still breaks?

Look closely at the PhysicsWallah article that passed.

5. Governance & Policy — CCPA fined PhysicsWallah ₹5 lakh for dark patterns. Small fine. Big signal. Every Indian growth team’s fake-urgency timer is now on legal notice. Industry L3(3) + Timely L4(8) + Revenue skipped + Teaching L5(20) = 31/35

Our AI model gave it a high score despite it not having a clear business learning. Why? The model used the rubric, which only said ‘broader market future prediction’ — and a single fine got inflated into a structural India signal. Vague language was producing a wrong answer.

Enter the calibration loop (human grader)
We updated the rubric. L5 now requires: a new law, regulation, or market structure shift. A single enforcement action does not qualify unless it creates a legal precedent that changes how all companies in the sector must operate.

After the update, the same story re-scored: L2.

Here’s what a good topic looks like now.

1. Media/Entertainment — Kuku: ₹240 Cr to ₹1,400 Cr in one year. AI cuts content cost. IPO filed. Microdrama is India’s next content category — AI built the moat, not the marketing budget. 🏷 🚩

Industry L4(4) + Timely L5(10) + Revenue L5(15) + Teaching L5(20) = 49/50

2. Sports Business — Zee got FIFA cheap. JioStar blinked at $15M. Zee locked 39 tournaments through 2034. In Indian media rights, patience beats a bigger cheque. 🚩

Industry L3(3) + Timely L5(10) + Revenue skipped + Teaching L5(20) = 33/35

This is just the tip of the iceberg.

We covered three graders here. A hard filter, a scoring rubric, and a calibration loop. That is a working eval for a newsletter topic scraper — built inside a Claude project, no code, no infrastructure. It is also the simplest version of what evals can do.

The same logic applies to customer support agents, research tools, coding assistants, onboarding flows or any system where you need to know whether the output is actually good before it reaches a user.

There are also eval types we haven’t covered, such as regression suites, capability benchmarks, and multi-trial consistency checks. Each solves a different failure mode.

We’ll walk through a few more in upcoming editions. Stick around for the ride.

Subscribe now

Master Hermes in 40 minutes.

Abhishek — Fri, 29 May 2026 12:11:45 GMT

You already think faster than you type. You’re just not using that ability for work.

Wispr Flow removes the friction of typing. With it, you can speak your thoughts and get clean, formatted text instantly. Our team uses it for everything from quick Slack replies to making strategy docs.

Sign up with GrowthX and get 3 months free.

Sign me up.

Today’s edition.

If you’ve used OpenClaw to set instructions that run routinely, you know OpenClaw is sort of terrible at retaining context. Persistent memory in OpenClaw requires manual setup, third-party plugins, and the right session configuration.

Hermes ships with memory on by default. It launched in February 2026 and crossed 140,000 GitHub stars in under three months because it does the one thing everyone wanted without asking: it remembers what happened last time.

Today, we’re using that to build a pricing monitor that runs periodically. No code. No server. No engineering ticket. Just plain English and 60 minutes.

But wait, why should we build a pricing monitor with Hermes?
A pricing monitor has three parts: a page reader, a comparison tool, and a logic that tells you when two versions of the page don’t match. Normally, that’s three separate tools, a database, and a developer.

Hermes collapses it all into one. It visits the page, holds the memory of what it saw last week, runs the comparison, and messages you on Telegram if anything moved. You don’t configure any of it. You describe what you want in plain English, and it figures out the rest.

Part 1: Install Hermes on your device.

1. Head to Hermes’ official website. Here’s the link.

This is what you should see.

2. Copy the install command.
3. Now, head to your Terminal and paste the install command. You should see a purple installer screen that says “Hermes Agent Installer”. This is the tool that installs Hermes onto your device. It takes about 4-5 minutes, so don’t close your window.

When it finishes, you’ll see a green Installation Complete screen with a list of next steps.

Part 2: Connect your AI model.

Hermes needs a brain. We’re using Google Gemini — free, no credit card, no expiry.

1. To get your Gemini API key, open aistudio.google.com in your browser and sign in with your Google account.

2. Click Get API Key in the left sidebar. Click Create API key, select a project from the dropdown, then click Create.
3. Copy the key somewhere safe and keep this tab open.

Run the setup wizard on your Terminal.
Copy the following exactly and hit Enter.

hermes setup

Next, select your AI provider.
A long list appears. Arrow down to Google AI Studio (Gemini models — native Gemini API) and hit Enter. Paste your API key when prompted, then press Enter. Then arrow down to gemini-2.5-flash and hit Enter.

The wizard will ask you to setup terminal backend next.
Keep current (local) selected. Hit Enter to confirm. This way, all commands run on your computer.

Now, set your working directory.
Hit Enter to accept the default [.] This is where Hermes sets commands to run periodic tasks.

Skip sudo.
Type N and hit Enter. This means Hermes cannot run administrator-level commands on your device.

Congratulations, Hermes now has a brain!

Heads up: Hermes runs on your device, so your laptop needs to be open and on for the monitor to work. It does not run when your Mac is shut down or asleep. If you want it running 24/7, you’ll need a cloud server. So, if you’re running a schedule, make sure you choose. time when you’re actually sitting on your laptop.

Checkpoint: If the installer gets stuck at “Trying SSH clone...”

Wait 60 seconds. If nothing moves, press Ctrl + C to cancel. Then run these two commands one by one:

git clone https://github.com/NousResearch/hermes-agent.git ~/.hermes/hermes-agent

cd ~/.hermes/hermes-agent && bash setup-hermes.sh

This does the same thing as the installer — just takes a different route.

When it finishes and asks “Would you like to run the setup wizard now?”, type Y and hit Enter.

Connect Telegram to Hermes.

In the Terminal, run this:

hermes setup gateway

A list of messaging platforms appears. Move to Telegram, get the Space button, and press Enter.

Paste your bot token when prompted and hit Enter.

How to get Telegram Bot token —
1. Open Telegram and search for BotFather
2. Type /newbot and follow the on-screen instructions.
3. Copy the HTTP API token.

Now move to the final steps.

Now open Telegram, search for @RawDataBot, and send it any message. Copy the number next to “id” in the reply. Come back to Terminal and paste it. Hit Enter.

When asked to use your user ID as the home channel, type Y and hit Enter.

When asked to install the gateway as a background service, type Y and hit Enter.

Congrats, your Telegram is now connected to Hermes!

Start the gateway to receive Telegram messages.

In the terminal, copy this exactly.

hermes gateway start

You’ll see Hermes confirm it’s running. It’s now listening for messages from your Telegram bot in the background.

Verify web search is on.

This will help our bot read pricing pages of the tools we mention on the web.

1. Run:

hermes setup tools

2. Select Configure CLI.
3. Scroll to the top of the list and confirm Web Search & Scraping has a [✓] next to it. If it does, hit ESC and move on. If it doesn’t, press Space to enable it, then Enter to confirm.

Part 4: Build the pricing monitor.

Everything is set up. Now you tell Hermes what to do — in plain English, on Telegram.

Open your Telegram bot. Send it this message, replacing the URLs with your actual competitor pricing pages:

Every Monday at 5pm, message me asking: ‘Any new pricing pages to add or remove this week? Reply ADD [url] to add, REMOVE [url] to remove, or SKIP to continue.’ Wait up to 12 hours for my reply. Every Tuesday at 9am, visit all URLs in the current list. On the first visit to any URL, save a snapshot of the full pricing page to a local file. On every subsequent visit, compare what you find against the saved snapshot. If anything changed — any plan, price, feature, or tier — send me a Telegram message explaining what changed and what it might mean. If nothing changed, send one message saying all clear. Start with this URL: https://notion.so/pricing

Hermes will confirm the schedule and the URLs with you. Read it. If anything looks wrong, correct it in plain English.

Test it immediately. Send this to your bot:

Run the pricing monitor now and send me the result.

Within 2 minutes, you should receive a Telegram message with everything Hermes found on the Notion pricing page.

If the message arrives, the monitor is live. Here’s my message for reference.

How to add or remove URLs.

The monitor starts with one URL. Adding more takes one message.

To add a competitor. Send your Telegram bot:

ADD https://[competitor pricing page URL]

Similarly, you can remove competitors or even check the list of competitors to track. Just send a message to your bot.

What’s next?

You just built something most teams don’t have. The same setup runs a changelog watcher, a job-posting tracker, and a review-spike detector. Same Hermes. Same Telegram. All you need to do is add a different prompt.

Found value in this edition? Share it with a friend who’d like it. It’s free!

How to build tools on top of Claude Chat

Abhishek — Tue, 26 May 2026 06:34:55 GMT

We’re coming to Mumbai Tech Week.
Speak Easy — Wispr Flow’s official after-hours at Mumbai Tech Week is being hosted with GrowthX. It’s a private room for a select group building, backing, and shaping the AI wave in India. Most guests are invited 1:1. A few spots are open. Want in?

Apply for access

Today’s edition.

Most people who use Claude or ChatGPT do so in the same way. They open a chat, type something, read the response, and close the tab. It’s a lot like how you’d talk to Google if it were a little bit ‘smarter’.

Claude also allows you to build apps on top of its chat interface with Instructions. Today, we’ll explore how to use this feature to build a Competitive Intelligence Brief Bot from scratch.

What are Instructions?

Claude allows you to set Instructions inside a Project. Think of Instructions as a rulebook — a fixed set of rules that the chat has to follow every time someone opens it.

That makes your Claude chat a workflow. The user has a text box to type into. Claude is the LLM layer that processes whatever comes in. The Instructions are the logic that connects the two. Put those three things together, and you have something that works a lot more like an app than a chatbot without writing a single line of code.

Part 1: Scoping the tool.

Here’s what a Competitive Intelligence Bot should have.

1. An input box to enter your own app to find reviews of similar tools
2. A way to get reviews
3. A way to analyse reviews

LLMs already serve as the analysis layer. And theoretically, you could use a scraper app to collect reviews, connect it to a form, and use an LLM to identify which tools to get reviews from.

The problem?
One, you’d get an extremely large response with hundreds of directions on what to build next. After all, G2 houses reviews for 1000s of apps. Besides, product managers already know which three or four tools are competing for the same customer they are trying to win. Those are the ones worth watching.

That made the scraper unnecessary.
G2 already generates a public RSS feed for any product’s review page. A product manager can open that feed in their browser, copy what is there, and paste it into Claude in ninety seconds. No scraper. No scheduler. No form. Just the reviews that matter, from the competitors that matter, processed immediately.

Part 2: Refining the tool.

Here’s the thing about reviews. They are an incredibly messy dataset.
They are written by people who felt strongly enough to spend ten minutes writing one, which means they are not a representative sample of all users. They skew toward extreme experiences. Someone who had a bad onboarding week writes about the entire product as though it is broken. Someone who got surprised by an invoice writes about pricing as though the company is dishonest.

Reviews also cannot tell you what the user tried before giving up, what they switched to after leaving, or whether the problem they are describing was fixed in a product update two weeks before they wrote about it. And because each reviewer is writing in isolation from every other reviewer, the same underlying problem can appear in ten different reviews described in ten completely different ways, making it look like ten separate problems when it is actually one.

The result is a dataset that is high volume, emotionally uneven, unrepresentative, lagging, and fragmented. Using it without any processing does not give you competitive intelligence. It gives you a list of complaints weighted by whoever happened to be angry enough to open G2 that week.

The fix?
We built a set of rules that run before any analysis happens. Think of them as filters that surface which data is worth paying attention to.

How do you know which rules to set?
Ask Claude. And keep testing the responses. Stop when you hit a passable response, with a good output format and the least tokens trifecta.

Here are the rules we settled on for our prompt:
1. Volume.
We limit the time window so that only recent reviews are included in the analysis. We set a minimum threshold so that a problem must appear more than once before it counts as a pattern. These two rules alone eliminate the majority of the noise.

2. Relevance.
We check each reviewer’s role and company size against the target customer's criteria before the review enters the analysis phase. After all, a review from the wrong customer is not a useful signal, regardless of how well-written it is.

3. Nature of the complaint itself.
Not every frustration represents a gap worth building for. We distinguish between a user who was annoyed and one who could not move forward, and only the second kind enters the brief. We also check whether the language suggests the user is actively evaluating alternatives, because that changes the urgency of the finding.

4. Honesty.
Every finding gets assessed against the PM’s own product to give a brief with relevant next steps. If the answer is unclear, the brief says so rather than defaulting to a reassuring conclusion.

None of these rules fixes every problem with reviews as a dataset. But together they narrow a large, uneven, emotionally driven corpus down to a small number of findings that are recent, relevant, and honest enough to be worth acting on.

Part 3: Instruction set up.

Once you land on an output you like, ask Claude to build instructions based on your conversation. Here’s the exact prompt you can paste into your chat to build instructions for your specific Competitive Intelligence Brief.

I have been refining a tool in this conversation. We have settled on the output format, the rules, and the workflow. I now need you to convert everything we have agreed on into a clean set of project instructions that I can paste directly into a new Claude project.

The instructions should cover three things in order.

First — the intake flow. Write it as a strict sequence where Claude asks one question at a time and waits for the user’s response before asking the next. Claude should never ask two questions at once. Each question should trigger the next only after the user has replied.

Second — the processing rules. Extract every rule, filter, and decision we agreed on during this conversation and write them as clear instructions Claude will follow every time the brief is run.

Third — the output format. Write the exact output structure we landed on, including the section headers, the card format, the field order, and the brief rules.

The instructions should be written so that a new Claude project with no prior context can run this tool correctly from the very first message — without referring back to this conversation.

Want the Instructions for the Brief we worked on?

When the user opens this project and sends their first message — including if they just say “hello”, “hi”, “ready”, or anything else — respond with exactly this and nothing else:

---

Welcome back. I am going to ask you four quick questions before we run your brief. Let’s go one at a time.

What is today’s date?

---

When the user replies with a date, respond with exactly this and nothing else:

---

Got it. Now I need to understand your product so I can map every finding against what you currently have.

Paste your product website or feature page URL here.

This tells me what your product does today, so I can tell you whether each competitor's problem is a gap you share, a gap you can claim, or something you have already solved.

---

When the user pastes their product URL, fetch it and read it. Use it as your reference point for the entire brief. Then respond with exactly this and nothing else:

---

Got it. Now for your competitor's feeds.

Paste the RSS feed content for your first competitor here.

To find it: go to G2, Capterra, or Trustpilot, search for your competitor, open their reviews page, and open one of these URLs in your browser:

- G2: `https://www.g2.com/products/[slug]/reviews.rss`

- Capterra: `https://www.capterra.com/p/[id]/[name]/reviews.rss`

- Trustpilot: `https://www.trustpilot.com/review/[domain]/rss`

Copy everything that appears and paste it here. Also tell me which platform it came from.

---

When the user pastes their first feed, respond with exactly this and nothing else:

---

Got it. Do you have another competitor feed to add?

If yes, paste it now in the same way.

If not, type “run brief”, and I will get started.

---

Repeat the same response each time the user pastes another feed, until they type “run brief.”

When the user types “run brief”, proceed with the analysis below. Do not show the user any of the processing steps. Return only the final brief.

How to run the brief

Step 1 — Identify the platform and normalise the format

Each feed will come from G2, Capterra, or Trustpilot. In all cases, extract only these three fields per review item:

pubDate
title (strip CDATA wrappers)
description (strip all HTML tags, bullet formatting, CDATA wrappers — extract plain text only from the substance of the review, ignoring star ratings, metadata, and reviewer credentials)

Discard everything else — links, GUIDs, XML tags, category metadata.

Step 2 — Filter by date
Use the date the user provided. Keep only reviews where pubDate falls within the last 7 days from that date. Drop everything older.

If fewer than 3 reviews qualify for any single competitor within the 7-day window, automatically extend the lookback to 30 days for that competitor only. Flag this in the brief so the product manager knows those signals are older data.

Step 3 — Tag each review with its source
Before analysis, prepend the competitor product name and platform to each qualifying review:

[Competitor name — G2] title — description text
[Competitor name — Capterra] title — description text
[Competitor name — Trustpilot] title — description text

Step 4 — Check signal strength
Only surface a problem if it appears in at least 2 reviews. A single mention goes into the Watch list only. Do not promote a single mention into a main finding.

Step 5 — Map each problem against the reference product
Using the product URL the user provided, determine one of three states for every problem identified:

Gap you share — your product has the same problem. Not an opportunity, a risk.
Gap you can claim — your product does not have this problem. Positioning and build opportunity.
Already solved — your product handles this. Communicate it louder.

If the product URL could not be fetched or returned insufficient information, tell the user before producing the brief and ask them to paste their product description directly into the chat instead.

Step 6 — Produce the brief
Use exactly this structure and format. All output must be in tables. No paragraphs anywhere. Caveman mode: short words, no fluff, only what matters.

COMPETITOR INTELLIGENCE BRIEF — [current week and date range] [Competitor 1 — Platform: X reviews in window] · [Competitor 2 — Platform: X reviews in window] Note any competitor where the 30-day fallback was used and why.

Main findings table

PriorityProblemWhoSignalJioMeet🔴 Critical[2–5 words. What users cannot do. No feature names.][Role · company size][Competitor: count · Platform · last 7d][Gap you share / Gap you can claim / Already solved — 1 punchy line on what to do]🟡 Highsame formatsamesamesame🟢 Watchsame format—[Competitor: 1 · Platform · date]Monitor / Awareness only

Rules for the main findings table:

🔴 Critical = problem in 3+ reviews or cross-competitor signal
🟡 High = problem in 2+ reviews
🟢 Watch = single mention only — no Who column, no JioMeet action beyond Monitor
Name the problem, not the feature. Never write “build X.” Write what users cannot do.
Never fabricate or infer a problem not directly supported by the reviews.
Never include direct quotes from reviews.
If a problem appears across multiple competitors, name all of them.

Best practices table

TopicDetailRun every[frequency based on review volume]Do this week[single highest priority action]Do today[any no-build action available]Can’t tell you[limitations of this brief]

Brief rules

Every cell must be scannable in under 2 seconds
No paragraphs anywhere in the output
No bullet points inside cells — use · to separate items inline
Signal expressed as: Competitor: count · Platform · timeframe
Watch list rows use — for Who and JioMeet columns
The entire brief must be scannable in under 2 minutes

Best practices for the product manager (for reference only — do not output these as prose)

Run frequency:

Competitor receiving 3+ reviews per day — run every 3 days
Competitor receiving 1–2 reviews per day — run weekly
Competitor receiving fewer than 3 reviews per week — run monthly

Platform selection: Use whichever platform has the most reviews for that specific competitor. Mix platforms across competitors.

How to act: For each Critical and High item, within 48 hours: open a discovery conversation, add to backlog with evidence, or explicitly dismiss with a written reason.

What this tool cannot tell you: Cannot confirm total reviews written that week — only what the RSS feed surfaced. Cannot access reviews behind a login. Cannot replace conversations with your own users.

What next?

We have shared the complete brief and the full instruction set that powers it. But the best way to understand what it does is to run it yourself. Paste the instructions into a new Claude project, drop in a competitor RSS feed, and see what comes out. The output will tell you more about what the tool is doing than any explanation we could write here.

From there, tweak. If the findings feel too broad, tighten the ICP. If the brief feels too long, cut a card format field. If the one question at the end does not feel specific enough, adjust the instruction that generates it. The instruction set is written in plain English precisely because it is meant to be edited by someone who is not a developer. Every rule in it is a decision that can be revisited once you see how the tool behaves in practice.

That’s all there is to it. Happy building!

Thanks for reading Build with AI! If you found this helpful, share it with a friend. It’s free :)

Build a Slack follow-up radar in 40 mins.

Abhishek — Thu, 14 May 2026 07:31:20 GMT

Most voice agents never leave demo. Solving cost, latency, and memory together is hard. Optimising it in a day — forget about it.

Up for the challenge? ElevenLabs is covering the credits. You pick the brain — OpenAI, Claude, or Gemini. We’ll give you the perfect focus room to build.

Sign me up. Now.

Today’s edition.

Your Slack’s going off all day. 100 notifications total. You need to focus so you hit Mark All as Read.

Boom. You just lost track of all your open threads.

Today, we’re building a Dependency Tracker for this exact issue. It scans your Slack workspace every 48 hours, finds messages you sent with no reply and direct mentions you haven’t responded to using AI, and posts a clean two-section summary straight to your Slack DM. All under 40 minutes.

Set a timer and open a new tab to view this email. We’re getting straight into it.

What you’ll need.
A Slack account with access to at least one workspace. Ideally, this should be the workspace where you could potentially have open threads. A free Groq account. This is our AI layer today, and it has a generous free tier. A Google Account. We’ll use it to test out code on Colab. And a Pipedream account (it’s similar to Make) to set up the actual scheduler and code.

Step 1: Get your Slack token

1. Go to api.slack.com/apps2. Click Create New App → From scratch

3. Name it anything — Dependency Tracker works. Select your workspace and click Create App

4. On the left sidebar, click OAuth & Permissions

5. Scroll to User Token Scopes and add every scope in this list:

channels:history
channels:read
groups:history
groups:read
im:history
im:read
im:write
mpim:history
mpim:read
search:read
chat:write

6. Scroll back up and click Install to Workspace
7. Copy the token that starts with xoxp- and save it somewhere safe

Step 2: Get your Groq API key

1. Go to console.groq.com

2. Sign up with your Google account — no credit card needed
3. Click the API Keys button at the top

4. Click Create API Key, give it any name, and click Submit

5. Copy the key and save it next to your Slack token

Step 3: Set up Google Colab (your test environment)

1. Go to colab.research.google.com

2. Click New Notebook and rename it — Dependency Tracker works
3. Click the key icon on the left sidebar to open Secrets
4. Add two secrets:

Name: SLACK_TOKEN — paste your xoxp- token as the value
Name: GROQ_API_KEY — paste your Groq key as the value

5. Toggle both secrets on

Step 4: Code testing in Colab.

1. Install the pip slack and groq libraries.
This installs the relevant Python packages for the Slack SDK and Groq so you can use them within Python. Here’s how you do it.
- Copy this line of code exactly into the first empty cell of your notebook and click on the small play button at the start of the cell.

pip install slack-sdk groq

You should see lines of text pop up on your screen and a small check mark next to your cell. This means your code ran successfully.

Next, add a new code block by clicking the Code button at the top.

Then, load your credentials. To do this, copy this code exactly into the new code block/cell and click on the play button to run the code.

from google.colab import userdata

SLACK_TOKEN = userdata.get(”SLACK_TOKEN”)

GROQ_API_KEY = userdata.get(”GROQ_API_KEY”)

If prompted, click Grant access. This pulls your tokens from the secrets panel into the script — they never appear in plain text in your code.

Step 4: Run the tracker.

Add a new cell by clicking the Code button, copy this code, and run it.

from slack_sdk import WebClient

from datetime import datetime, timedelta

from groq import Groq

import time

slack = WebClient(token=SLACK_TOKEN)

groq_client = Groq(api_key=GROQ_API_KEY)

my_user_id = slack.auth_test()[”user_id”]

cutoff = datetime.now() - timedelta(days=2)

cutoff_ts = cutoff.timestamp()

channels = slack.conversations_list(

types=”public_channel,private_channel,im,mpim”,

limit=200

)[”channels”]

section1 = []

section2 = []

for channel in channels:

channel_id = channel[”id”]

channel_name = channel.get(”name”, “DM”)

try:

messages = slack.conversations_history(

channel=channel_id,

oldest=cutoff_ts,

limit=100

)[”messages”]

for msg in messages:

if msg.get(”bot_id”) or msg.get(”subtype”) == “bot_message”:

continue

msg_ts = float(msg.get(”ts”, 0))

days_waiting = (datetime.now() - datetime.fromtimestamp(msg_ts)).days

if days_waiting < 2:

continue

thread_ts = msg.get(”thread_ts”) or msg.get(”ts”)

try:

replies = slack.conversations_replies(

channel=channel_id,

ts=thread_ts

)[”messages”]

except:

replies = [msg]

human_replies = [

r for r in replies

if not r.get(”bot_id”) and r.get(”subtype”) != “bot_message”

]

last_sender = human_replies[-1].get(”user”) if human_replies else None

if msg.get(”user”) == my_user_id and last_sender == my_user_id:

section1.append({

“channel”: channel_name,

“message”: msg.get(”text”, “”)[:120],

“days_waiting”: days_waiting

})

if f”<@{my_user_id}>” in msg.get(”text”, “”) and msg.get(”user”) != my_user_id:

if msg.get(”bot_id”):

continue

i_replied = any(

r.get(”user”) == my_user_id

for r in human_replies[1:]

)

if not i_replied:

section2.append({

“channel”: channel_name,

“message”: msg.get(”text”, “”)[:120],

“days_waiting”: days_waiting,

“status”: “never replied”

})

else:

last_messages = “\n”.join(

[r.get(”text”, “”) for r in human_replies[-3:]]

)

prompt = f”“”Read these Slack messages and answer only YES or NO.

Is this conversation still unresolved or waiting for further action?

Messages:

{last_messages}

Answer only YES or NO:”“”

response = groq_client.chat.completions.create(

model=”llama-3.3-70b-versatile”,

messages=[{”role”: “user”, “content”: prompt}],

max_tokens=10

)

answer = response.choices[0].message.content.strip().upper()

if “YES” in answer:

section2.append({

“channel”: channel_name,

“message”: msg.get(”text”, “”)[:120],

“days_waiting”: days_waiting,

“status”: “unresolved”

})

time.sleep(0.3)

except:

continue

section1.sort(key=lambda x: x[”days_waiting”], reverse=True)

section2.sort(key=lambda x: x[”days_waiting”], reverse=True)

output = “*🔴 Dependency Tracker*\n\n”

output += “*You are waiting on someone:*\n”

if section1:

for item in section1:

output += f”• #{item[’channel’]} — {item[’message’]} _({item[’days_waiting’]} days ago)_\n”

else:

output += “• Nothing. All clear.\n”

output += “\n*Someone is waiting on you:*\n”

if section2:

for item in section2:

output += f”• #{item[’channel’]} — {item[’message’]} _({item[’days_waiting’]} days ago, {item[’status’]})_\n”

else:

output += “• Nothing. All clear.\n”

dm = slack.conversations_open(users=my_user_id)

dm_channel_id = dm[”channel”][”id”]

slack.chat_postMessage(channel=dm_channel_id, text=output)

print(”Posted to your Slack DM.”)

print(”\nPreview:”)

print(output)

When it finishes the run, you should see ‘Posted to your Slack DM ‘ in the notebook, and the summary arrive in your Slack DM.

Step 5: Automate your code with Pipedream

1. Go to pipedream.com and sign up with your Google account — no credit card needed.
2. Click New Workflow and name it Dependency Tracker. Click on Create project and continue.

3. Set the Timeout to 300 seconds and click Create Workflow. Doing this gives enough time for your script to go through all your Slack messages and highlight conversations that need to be followed up on. Don’t worry, this 300s lag will happen in the background, so you’ll never really notice it.

4. You should now see your workspace with an Add Trigger button. Click on it.

5. Select Schedule, then click the Custom interval tab.

6. Set the interval to every 48 hours. This ensures your script runs only once every 2 days. Then set your time zone to your location.

7. Click the + below the trigger and select Run Python code.

8. Next, click on the Cube at the top of your screen. Go to Environment Variables in the left sidebar and add two variables:

SLACK_TOKEN — paste your xoxp- token
GROQ_API_KEY — paste your Groq key

9. Now, click on Dependency Tracker, navigate to the code step and click on it. Clear the editor, and paste the code below.

import os

from slack_sdk import WebClient

from datetime import datetime, timedelta

from groq import Groq

import time

def handler(pd: “pipedream”):

SLACK_TOKEN = os.environ.get(”SLACK_TOKEN”)

GROQ_API_KEY = os.environ.get(”GROQ_API_KEY”)

slack = WebClient(token=SLACK_TOKEN)

groq_client = Groq(api_key=GROQ_API_KEY)

my_user_id = slack.auth_test()[”user_id”]

cutoff = datetime.now() - timedelta(days=2)

cutoff_ts = cutoff.timestamp()

channels = slack.conversations_list(

types=”public_channel,private_channel,im,mpim”,

limit=200

)[”channels”]

section1 = []

section2 = []

for channel in channels:

channel_id = channel[”id”]

channel_name = channel.get(”name”, “DM”)

try:

messages = slack.conversations_history(

channel=channel_id,

oldest=cutoff_ts,

limit=100

)[”messages”]

for msg in messages:

if msg.get(”bot_id”) or msg.get(”subtype”) == “bot_message”:

continue

msg_ts = float(msg.get(”ts”, 0))

days_waiting = (datetime.now() - datetime.fromtimestamp(msg_ts)).days

if days_waiting < 2:

continue

thread_ts = msg.get(”thread_ts”) or msg.get(”ts”)

try:

replies = slack.conversations_replies(

channel=channel_id,

ts=thread_ts

)[”messages”]

except:

replies = [msg]

human_replies = [

r for r in replies

if not r.get(”bot_id”) and r.get(”subtype”) != “bot_message”

]

last_sender = human_replies[-1].get(”user”) if human_replies else None

if msg.get(”user”) == my_user_id and last_sender == my_user_id:

section1.append({

“channel”: channel_name,

“message”: msg.get(”text”, “”)[:120],

“days_waiting”: days_waiting

})

if f”<@{my_user_id}>” in msg.get(”text”, “”) and msg.get(”user”) != my_user_id:

if msg.get(”bot_id”):

continue

i_replied = any(

r.get(”user”) == my_user_id

for r in human_replies[1:]

)

if not i_replied:

section2.append({

“channel”: channel_name,

“message”: msg.get(”text”, “”)[:120],

“days_waiting”: days_waiting,

“status”: “never replied”

})

else:

last_messages = “\n”.join(

[r.get(”text”, “”) for r in human_replies[-3:]]

)

prompt = f”“”Read these Slack messages and answer only YES or NO.

Is this conversation still unresolved or waiting for further action?

Messages:

{last_messages}

Answer only YES or NO:”“”

response = groq_client.chat.completions.create(

model=”llama-3.3-70b-versatile”,

messages=[{”role”: “user”, “content”: prompt}],

max_tokens=10

)

answer = response.choices[0].message.content.strip().upper()

if “YES” in answer:

section2.append({

“channel”: channel_name,

“message”: msg.get(”text”, “”)[:120],

“days_waiting”: days_waiting,

“status”: “unresolved”

})

time.sleep(0.3)

except:

continue

section1.sort(key=lambda x: x[”days_waiting”], reverse=True)

section2.sort(key=lambda x: x[”days_waiting”], reverse=True)

output = “*🔴 Dependency Tracker*\n\n”

output += “*You are waiting on someone:*\n”

if section1:

for item in section1:

output += f”• #{item[’channel’]} — {item[’message’]} _({item[’days_waiting’]} days ago)_\n”

else:

output += “• Nothing. All clear.\n”

output += “\n*Someone is waiting on you:*\n”

if section2:

for item in section2:

output += f”• #{item[’channel’]} — {item[’message’]} _({item[’days_waiting’]} days ago, {item[’status’]})_\n”

else:

output += “• Nothing. All clear.\n”

dm = slack.conversations_open(users=my_user_id)

dm_channel_id = dm[”channel”][”id”]

slack.chat_postMessage(channel=dm_channel_id, text=output)

return {”status”: “sent”}

Click Test — you will get a Slack DM confirming it works. Then, click Deploy.

Try it and tell us what you think!

Thanks for reading Build with AI! If you like this post, share it with a paid ads marketer who’d like this edition. It’s free :)

Building Meta Ads with Claude

Abhishek — Thu, 07 May 2026 05:31:06 GMT

You’ve watched the tutorials. You’ve tried to follow along. Then the tutor says “just pip install the library” — and suddenly you’re in a whole new rabbit hole.

Building with AI is hard when you’re starting out. We make it easier by putting you in the same room as builders at the frontier — online and offline.

Check out GrowthX events

Today’s edition.

One week ago, Meta launched its official Meta Ads Model Context Protocol (MCP) server for Claude. And if you’re a marketer, that means your life just got a whole lot easier. Now, thinking, analysis, and creative direction can happen from within Claude.

Today, we’ll connect the Meta Ads MCP to Claude and build a simple ad validator that checks your new creatives against the patterns in your best-performing ads — and tells you exactly what to fix before you spend a single rupee on them in 60 minutes.

Set a timer and open a new tab. We’re getting straight into it.

What you’ll need.
A Claude Pro account (~ ₹1,900/month — use claude.ai on the web, not the app store, to avoid extra charges). You can access MCP only on a Pro or Max plan today. We recommend starting with the Pro. Your signed-in Meta Ads account has a few past creatives. That’s it.

Heads up: This works if you’ve got the Meta Ads MCP update only.

Step 1: Setting up the Meta Ads MCP

The official Meta Ads MCP endpoint is mcp.facebook.com/ads. Here is how to add it to Claude.

1. Open Claude — desktop app or claude.ai in your browser.
2. Go to Settings → Integrations (on Claude.ai) or Settings → Connectors (on Claude Desktop).
3. Click Add Integration and paste this URL. Enter the URL as shown and click Add.

4. You’ll see the Meta Ads MCP get added to your connectors. Click the ‘Connect’ button. Claude will prompt you to log in with your Meta Business account — the same login you use for Ads Manager. Click through and authorise access. Here’s what you should see.

Congratulations! You’ve connected successfully to Meta Ads MCP.

Step 2: Set up your Claude Project

A Claude Project is a workspace that remembers your brand permanently. Every time you open it, Claude already knows who you are, what you sell, and exactly what to ask you. You set this up once. You never repeat it.

Step 2a: Create a new Project.
Open Claude.
1. In the left sidebar, click Projects → New Project.

2. Name it something you will recognise — Ad Validator works perfectly.
3. You will see a field called Project instructions. Click on it. This is where the rulebook lives.

4. Head over to instructions and copy this prompt into your project.

You are an ad creative validator for [Brand Name].

Every time a new chat starts in this Project, begin by saying:

“Let’s build your creative direction. I need to understand your current campaign first. Please answer these questions one at a time. I have included an example answer under each question to show you the level of detail I need.”

Then ask these seven questions one at a time, waiting for the answer before moving to the next:

1. What are you promoting right now — specifically, not generally?

Example: “We are promoting our June AI cohort. It is a 6-week live program, ₹25,000, starting June 9th. 50 seats total.”

2. Is there a deadline or urgency attached to this campaign? If yes, when exactly?

Example: “Early bird pricing closes May 31st. After that the price goes up by ₹5,000.”

3. What is the one action you want someone to take after seeing this ad?

Example: “Click through to the landing page and fill in the interest form. Not buy yet — just express interest.”

4. What messaging have you already run in this campaign that the audience has already seen?

Example: “We have already run ads about the curriculum, the instructors, and the career outcomes. The audience has seen all three angles.”

5. Is this audience cold — seeing you for the first time — or warm — they already know your brand?

Example: “Warm. This is a retargeting audience — people who visited our landing page in the last 30 days but did not fill in the form.”

6. What is the one objection or fear stopping this audience from clicking right now?

Example: “They are probably thinking — I am too busy right now, I will do it next time. Or — I am not sure if this is worth ₹25,000 for someone at my level.”

7. How do you want this person to feel after they click?

Example: “Like they are about to miss something that everyone around them will have done in six months and they will not.”

Once you have all seven answers, say: “Got it. Let me now check your Meta Ads account for live campaign data.”

Then use the Meta Ads MCP to check whether any campaigns are currently active and spending.

IF LIVE CAMPAIGNS ARE FOUND:

Use the Meta Ads MCP to pull performance trend data on the active campaigns. Identify the three best-performing ads by CTR and the three worst-performing ads by CTR. Exclude any ad with fewer than 1,000 impressions — CTR on low impression counts is statistically meaningless.

For the worst performing ads, note when performance started dropping and which metric dropped first — CTR, CPM, or ROAS. Then skip to the analysis section below.

IF NO LIVE CAMPAIGNS ARE FOUND:

Do not tell the user everything at once. Guide them through Ads Manager one step at a time. After each instruction, ask them to confirm what they see on screen before giving the next step. Use the following flow exactly:

Say: “There are no live campaigns running right now, so I cannot pull your data automatically. I am going to walk you through finding your historical ads in Ads Manager. We will do this one step at a time. Tell me what you see after each step, and I will guide you to the next one.

First step:
Open a new tab and go to facebook.com/adsmanager. Once it loads, tell me what you see on the screen.”

Wait for their response. Based on what they describe, do one of the following:
1. If they say they can see a dashboard with campaigns listed, say: “Perfect, you are in the right place. Now look at the top of the page. You should see three tabs next to each other — one that says Campaigns, one that says Ad sets, and one that says Ads. Can you see those three tabs? Click the one that says Ads and tell me what appears.”

2. If they say they are being asked to log in, say: “You need to log in first. Use the Facebook account that has access to your ad account. Once you are logged in and can see a dashboard, tell me what you see.”

3. If they say they see something else, ask them to describe it or take a screenshot and share it. Use what they share to redirect them correctly.

Second step:
Once they confirm they are on the Ads tab, say: “Good. Now look at the top right corner of the table — not the top right of the whole browser, but the top right area just above the list of ads. You should see a date selector that

probably says something like Last 30 days or a specific date range. Do you see it? Click on it and tell me what options appear.”

Wait for their response. Once they confirm they can see date options, say: “Look for an option called Maximum on the left side of the date picker. It may be in a list of presets like Today, Yesterday, Last 7 days, and so on. Click Maximum and then click the Update or Apply button. This will show you every ad that has ever run on this account. Tell me what you see now — roughly how many ads appear in the list?”

Wait for their response. Then say: “Now we need to sort these ads by CTR so we can find your best and worst performers. Look at the column headers at the top of the table — the row that says things like Ad name, Delivery, Results, and so on. Scroll right along those headers until you find one that says CTR or Link click-through rate. Can you see it? Click that column header once. This will sort all your ads from highest CTR to lowest. Tell me what the top three ads are called and what their CTR percentages are.”

Wait for their response. Note the names and CTR numbers of the top three ads. Then say: “Great. Now I need you to find the creative — the actual image or video — for each of those three ads. Click on the name of the first ad in the list.

A panel should open on the right side of the screen, or it may take you to a new page. Look for a tab or button that says Preview. Click it. You should now see the ad as it appeared to your audience. Can you see the image or video? Take a screenshot of it and upload it here.”

Wait for them to upload the first creative. Then repeat the same instruction for the second and third best performing ads, one at a time.

Once you have all three best performing creatives, say: “Perfect. Now I need the three worst performing ads. Go back to the full ads table. Click the CTR column header again — this time it will flip the sort so the lowest CTR ads appear at the top. These are your worst performers. Tell me what the top three ads are called and what their CTR percentages are.”

Wait for their response. Note the names and CTR numbers. Then guide them through finding and uploading the creatives for each of the three worst performing ads, one at a time, using the same Preview method as above.

Once you have all six creatives and all six CTR numbers, say: “I now have everything I need. Let me run the analysis.”

ONCE YOU HAVE EITHER LIVE DATA OR UPLOADED CREATIVES:
Determine which situation applies:

SITUATION A — The ad performed poorly from day one.

This is a concept problem. The creative direction was wrong from the start. The fix is a completely new creative concept.

SITUATION B — The ad performed well for 2 or more weeks and then dropped.
This is a fatigue problem. The concept was right but the audience has seen it too many times. The fix is a fresh execution of the same idea — not a new concept.

State clearly which situation applies and why, in one short paragraph, before giving the output.

Then apply these seven visual questions to the three best performing ads:

1. Is the imagery real or AI-generated?
2. Where is the human in the frame?
3. Are faces shown?
4. What is the colour temperature?
5. How does text sit on the image?
6. What feeling does the image produce before you read a word?
7. What is notably absent?

Identify the visual pattern across the three best performers.

Apply the same seven questions to the underperforming ads.
Identify specifically what the underperforming ads are missing compared to the pattern.

Then produce the output in this exact structure:

SITUATION: state A or B and one sentence explaining why.
DIAGNOSIS: two to three sentences in plain English explaining why the underperforming ads failed and what the best performers did differently.

AD COPY OPTION 1:
Hook — under 8 words, no punctuation at the end
Body — one sentence, specific promise
CTA — action-first, under 5 words

AD COPY OPTION 2:
Hook — under 8 words, no punctuation at the end
Body — one sentence, specific promise
CTA — action-first, under 5 words

CREATIVE DIRECTION ANGLE 1 — give it a name:
Two to three sentences describing the visual concept. What is the scene. Where is the human. What are they doing. What is impossible or unexpected about the setting. What feeling does it produce before any text is read.

Image generation prompt: Write a complete prompt that can be pasted directly into any AI image model such as Midjourney, DALL-E, or Firefly. The prompt must describe the subject, the setting, the lighting, the colour temperature, the position of the human in the frame, whether the face is shown, what the mood is, and what is deliberately absent.

Write it as one continuous paragraph of three to five sentences. Do not use bullet points inside the prompt. Make it specific enough that two different people using it would produce visually similar results.

CREATIVE DIRECTION ANGLE 2 — give it a name:
Same structure as Angle 1.
Image generation prompt: Same structure as Angle 1.

CREATIVE DIRECTION ANGLE 3 — give it a name:
Same structure as Angle 1.
Image generation prompt: Same structure as Angle 1.

COPY RULES:
The tone must be direct, aspirational, and specific. Never generic. Never safe. The language must feel like it was written by someone who understands what it means to be a working professional trying to stay relevant in an AI-first world.

Never use the words: transformative, game-changing, unlock, elevate, journey, empower, or leverage.

IMAGE PROMPT RULES:
Every image generation prompt must follow the visual pattern identified from the best performing ads. It must specify whether the image is real-feeling or AI-generated in style, where the human is placed in the frame, whether faces are shown and what expression if so, the colour temperature and mood, how text should sit on the image, the emotional feeling before any text is read, and what is deliberately left out of the frame. Write each prompt as a single flowing paragraph — not a list. Make it specific enough that two different people using it would produce visually similar results.

IMPORTANT TECHNICAL NOTE:
The Meta Ads MCP connected to this Project uses Meta’s official MCP server at mcp.facebook.com/ads. This server can only surface data from campaigns that are currently active and spending. It cannot retrieve historical performance data, creative assets, or CTR numbers from paused or completed campaigns. If all campaigns are off, do not attempt repeated MCP queries. Go directly to the step-by-step Ads Manager guidance above.

5. Once you start a new chat inside the Project, Claude will ask you seven questions about your current campaign — one at a time. Answer them as specifically as you can. The more detail you give, the sharper the output. Claude will then check your Meta Ads account for live data. If your campaigns are active, it pulls everything automatically. If not, it walks you through finding your historical ads in Ads Manager step by step.

At the end, you will have three things: a plain English diagnosis of why your ad is not working, two ad copy options ready to use, and three image generation prompts you can paste directly into Gemini or OpenAI to produce your next creative. The whole process takes only ~15 minutes once the setup is done.

Try it, and tell us what you think.

Thanks for reading Build with AI! If you like this post, share it with a paid ads marketer who’d like this edition. It’s free :)