Where product marketers are struggling to create value and influence, I feel like it’s the opposite with competitive intelligence (CI) folks.

Maybe it’s because PMMs often create internal facing assets (positioning canvas, messaging docs, battlecards, etc) but when you can track what market leaders are doing and leveraging that for your new campaign ideas, it’s usually a good way to show face value.

However, the last thing we need as product marketers is more stuff to add to our todo list.

That’s why today, I’m showing you how you can build an agent that is helping you spy on the competition, without the burden of researching it yourself.

If you haven’t built anything with AI yet, this is a fun project you can finish in ~ 15 minutes. Let’s get started 👇

Stop typing what you could say in 10 seconds.

Wispr Flow turns your voice into clean, professional text inside any app. Emails, Slack, client updates — speak once, send without editing. 4x faster than typing.

This is an ad, but I’ve been using Wispr Flow for the past few months and it’s a GAME-CHANGER to the way I’m using AI.

It’s way faster than typing, I can chat with my friends just by talking for a few seconds, and it copies my own style, in FR & EN. Highly recommended

The tools you need: Firehose + Claude

Firehose is a real-time data streaming API. In plain English, it means that when there’s specific words or filters that are being identified, it will appear in Firehose.

It’s built by ahrefs, and I’ve been using it to receive daily updates on AI news for PMMs and GTM + tracking the competition last moves for Guideflow.

Here’s how you can create an daily agent that will give you email or Slack notification about the competition

1. Define the scope of your CI program

Don't try to track "the market." Pick the box.

For Guideflow, I picked 12 named competitors in the interactive demo space. Navattic, Storylane, Walnut, Tourial, and 8 others.

Then I picked what counts as a signal: Funding, Product launches, Partnerships, Press, Hiring, Pricing. Anything outside that goes into a "General" bucket I check weekly.

If you can't write your competitor list and signal types in 5 minutes, you don't have a scope. You have a vibe.

2. Create your first tap

A "tap" is just Firehose's word for a stream.

Log in, create one, name it after the program (mine is "Guideflow CI Monitoring").

One CI program = one tap. Don't split.

3. Add rules and language exclusion

Twelve competitors = twelve rules. Each one is a Lucene query (mini-language to retrieve terms on the web)

Easy brands: (navattic) OR title:navattic. Done.

Tricky brands (where the name is also a regular word): add domain and product hints. For Reprise, my rule is (reprise OR "reprise demo") OR title:reprise. Without it, I just get garbage.

For language, I don't put AND language:"en" in the rule. Brand names are universal, so filtering at the rule level kills recall. I filter in the Python script after the events come in.

Cap is 25 rules per org so twelve brands leaves you room.

4. Share taps and management keys with Claude

Put the keys in a .env file next to your script. FIREHOSE_TAP_TOKEN, FIREHOSE_MGMT_KEY, and your delivery key (Slack webhook for me).

Add .env to your .gitignore BEFORE you paste anything. Don't be that guy. Otherwise you’re creating a security risk.

Then drop the Firehose API doc into the project as a markdown file. I save it as firehose-api.md. Claude reads it once and now knows every endpoint.

That's the moment Claude stops being a chatbot and starts being your engineer.

5. Build the context of this agent

This is the part nobody talks about, and it's where 90% of the value lives.

For brand monitoring, "context" isn't a fancy prompt. It's three things:

→ Your competitor map (brand + Lucene query + tag)
→ Your signal classifier (e.g. a dict mapping keywords to categories, like "raised, series, valuation" → Funding)
→ Your matching logic (use the rule tag first, fall back to content match, never let an event drop silently)

Add a freshness cutoff. Mine is 14 days. Old funding rounds aren't news.

And write your empty-state message NOW, before launch. Mine says "All quiet on the western front." If the channel goes silent without one, the team will assume the bot broke.

6. Pick the best channel for daily updates

The channel is part of the product.

For Guideflow: Slack. The GTM team lives there. Email would die in their inbox.

I use Slack Block Kit (header, date, one section per competitor, emoji per signal type, footer with totals). Mobile-readable, scannable in 10 seconds.

Match the channel to where your reader already is. Don't make them come to you.

7. Iterate and improve based on results

CI programs decay faster than other automations. Build the loop in.

Every week, check three things:

→ Is the "Unclassified" bucket growing? Your rules are too broad or too narrow.
→ Which signal types did we actually act on? Cut the ones that never move the needle.
→ Did any competitor pivot or rebrand? Update the rule.

I have an update-rules command that wipes and recreates everything from the config. Two minutes. The iteration loop lives in code, not on a calendar reminder.

Unsure where to start? I gotchu.

Here’s a video on the exact steps as a walkthrough in 6 minutes:

If there’s anything that is not clear, feel free to drop a comments!

Why CI is a low hanging fruit to build with AI

I use to admire competitive intel from afar and thought it was so cool. But then, I looked at my Asana dashboard and was like “no bueno”. Can’t spare anything else towards this.

Building something that tracks your competitors movement is important because it shows that you are a proactive PMM, that also knows their ways around bots. Especially if you’re starting out, it can give out campaign ideas, help supporting the roadmap, and overall just be more aware of what is happening in your market.

So if you’ve been looking for a quick side project to show how cool and smart you are, try one of these two methods.

Your future self (and hopefully your boss) will thank you.

That’s all for this week, salut 👋

The pre-AI PMM only has 18 months to live

If you have not realized it yet, there’s a massive disruption happening for product marketers in B2B SaaS. In 6 months, we will see a clear separation between PMMs that 5x their outputs vs. the ones that are still tinkering with AI.

In 18 months, I bet that only a few laggards will still work as we all used to. There’s a clear zeitgeist happening and if you don’t follow the curve, you’re behind.

I’m starting filling out my waitlist for May to work with me. The offer is simple:

I build your employee zero
before one replaces you

You bring the product, the messaging, the positioning. I build the AI system that runs on all of it. And makes your work visible while feeling like a cheat code.

By Friday, it's how your team sees what product marketing actually moves.

Keep Reading