Good morning.

There's a version of competitor monitoring that runs perfectly but is worth nothing:

It fires every Monday, the findings are accurate, the summary is well written, but nobody has changed a single decision because of it in eight months.

Today I'll show you the agent that avoids that, why the fix is a judgment layer rather than a better scraper, and the one change that turns a competitor report into a gap report.

— Sam

Let’s get into it.

An agency owner I work with had competitor monitoring running before we ever talked about it.

Good setup: five competitors, weekly scan, a tidy digest in his inbox every Monday at eight.

When I asked what he'd done differently because of it, he scrolled back through about fifteen reports and found one thing, and it was a link he'd forwarded to someone.

The reports weren't wrong. They said things like Competitor B published four new case studies and Competitor D updated their homepage headline, and each of those was true.

What none of them said is whether any of it should cost him ten minutes of his week. So he read them the way you read a newsletter you like but don't need, and he stopped opening them around week nine.

That's the standard outcome, and it comes from an agent built to detect changes rather than to rule on them. Spotting the change is the easy half, and it's the half every tool sells you.

What an entrepreneur needs is the second half. Does the change move what a buyer sees when they're choosing between you and them, and what do you owe in response?

There's a second problem underneath that one, and it's the more expensive of the two. Competitor monitoring watches outward. You get a detailed picture of what four other businesses are doing and no picture at all of how you look next to them.

So the report can tell you a competitor added a mid-tier at $490. It has nothing to say about your own mid-tier, which has been sitting at $850 with a thinner deliverable since March. The gap is the part you can act on, and a watch pointed only at them never sees it.

The agent in this issue fixes both. It watches a fixed set of claims rather than a list of pages. It puts your own business on that same sheet, so every change it finds reads as a gap.

Then it rules on each difference, which is what makes the report worth opening.

Two paths, and the choice comes down to where the rest of your work already lives.

Claude Code in Cowork mode suits you if you want the agent reading and writing real files on your machine, keeping the claim sheet as a document you can open, and running as a Skill you invoke by name. The claim sheet is a file, the history is a folder, and you can read both without asking the agent anything.

Codex suits you if your business runs inside a browser and you'd rather the agent live where your other automations already do. No Skill primitive, so the configuration is a system prompt plus a scheduled run, and the claim sheet lives in whatever document store you already use.

The quick decision rule:

Both produce the same report, and the architecture below doesn't change between them.

The architecture has two halves. The Claim Sheet is what the agent watches. The Verdict is what it does with a change.

The watch is pointed at claims rather than at pages. A page gets rebuilt every couple of years and the claim underneath it usually survives.

A claim is a thing a business says about itself, and it stays the same thing whether it sits on a pricing page, in a footer, or halfway down a sales letter.

Five slots per competitor, and these are the five a buyer weighs when they're deciding between you and someone else:

That's the whole sheet: five slots against three to five competitors.

Hiring and ad creative don't get slots of their own, and the reason matters: a job posting is a leading indicator of a claim rather than a claim itself. Three enterprise AE roles opening says their Buyer slot is about to move, and four backend engineers says the same about the Promise.

The agent reads hiring pages and ad libraries as evidence feeding the five slots. Neither gets a paragraph of its own, and the report stays on what a buyer will see.

You go on the sheet too, in the same five slots, filled with the same rigor. This is the change that does the most work in the whole build.

Your row gets refreshed on the same schedule and from your own live pages, because entrepreneurs are consistently wrong about their own claims.

Usually the pricing page still says something that stopped being true two quarters ago. An agent reading your site cold will catch that on the first run.

Once your row exists, every competitor change resolves against it automatically.

Competitor C dropped their entry tier to $290 stops being news and becomes Competitor C's entry tier is now $290 against your $450, and their entry tier includes the onboarding call yours charges for.

The second version tells you what a prospect comparing two tabs is looking at.

Every difference the agent finds gets exactly one of four rulings, and the rulings are the reason the report gets read.

Three disciplines hold the whole thing together, and dropping any one of them turns the agent back into the report nobody reads.

It has to be able to tell you that nothing happened. An agent that finds something every single run is an agent that's promoting Noise to fill space, and that is exactly how a report loses its reader. Four changes detected, all Noise, nothing to answer this week is a valid and useful output, and after a month of them you'll have learned something real about how fast your market moves.

Every claim has its source and the date it was read. A claim without a URL and a timestamp can't be checked, and one wrong price in week three costs you the credibility of the whole sheet. Anything the agent can't verify against a live page gets marked unconfirmed and stays out of the verdict column rather than getting asserted.

It never rules Move on its own authority and it never recommends a price. The agent flags that a pricing decision is now live and puts it in front of you with the gap attached. Pricing is the single fastest way to do real damage on bad information, and an agent working from a competitor's public page is working from marketing rather than from their margins.

Six steps. The prompts are complete, and the only thing you're filling in is your own business.

Three to five, and the selection rule matters more than the number. Pick the businesses that show up in the deals you lose.

The market leader you've never once been compared against is a distraction with a good logo. Every slot it takes is a slot not spent on the outfit that beat you on two proposals last quarter.

If you're not sure, the answer is already in your own records. Pull the last twenty closed-lost deals and count the names that appear. Ask whoever runs your sales calls who comes up. That list is your watch set.

This is the file the agent reads on every run, and the quality of it decides the quality of everything downstream. Fill it in properly once and you'll edit it twice a year.

The two sections entrepreneurs skip are What Counts As Material To Me and My Live Constraints, and skipping them produces a report full of Noise dressed as findings.

The first one teaches the agent your judgment. The second stops it handing you six Answers you have no way to act on this quarter.

The first run reads every page in the context file and fills the sheet cold. It rules on nothing, because there's nothing to compare against yet.

Save the output as claim-sheet-[date].md. That file is now the thing every future run compares against.

This is the run that repeats. It refreshes the whole sheet, compares against the stored one, and rules on every difference.

Weekly, and run it on a day that gives you room to act rather than on Monday morning when your week is already committed.

The agency owner I mentioned moved his to Thursday and started clearing Answer items on Friday afternoons. That was the first week his monitoring changed anything.

In Cowork, this is a scheduled Skill invocation against the context file. In Codex, it's a scheduled run with the same prompt and the sheet path pointed at your document store.

Either way, the run needs write access to wherever the sheets are kept, because the value compounds in the history rather than in any single report.

Monthly is too slow for price and packaging, which is where the expensive surprises live. Daily produces a stream of Noise and trains you to stop reading, which is the failure this whole architecture exists to prevent.

An Answer with no owner and no date is just a Watch. When the report gives you two Answer items, put a name and a date against each one before you close the tab, and keep them somewhere you'll see them again.

Most Answers are small, and usually one of three things:

Each one closes a gap a prospect was going to find on their own.

The compounding version is a business whose public claims stay current against the market every week, with nobody sitting down to do a quarterly positioning review.

Agency. The watch set is the three shops that come up in your proposals. The Price slot is where most of the action is, because agency pricing is opaque and a competitor publishing a package price for the first time is an event.

Path matters more here than people expect: an agency moving from book a call to a published productized tier has changed how buyers evaluate the whole category, and your own Path row is what tells you whether that leaves you looking considered or looking slow.

SaaS. Price and Proof do most of the work here. Tier restructuring and per-seat versus usage changes are the differences that decide comparison shopping, and the Proof slot catches the moment a competitor wins a logo in your core segment.

Careers pages are the highest-value input, because a hiring run against a new segment shows up in the report as a Watch about six months before the positioning changes.

Ecommerce. Price includes shipping thresholds, bundle composition and returns terms, which is where most of the real movement happens.

Proof means review counts and rating, and those move on their own without anyone deciding anything, so set the material threshold in the context file at a change of a certain size, which keeps ordinary drift out of the report.

Expert, consultant, creator. Promise and Proof are almost the whole game, because the offers are close and the differentiation is in what's claimed and what's shown.

The most useful thing the mirror row does for this reader is catch that a competitor's Promise has become more specific than theirs, which is a gap that costs deals long before anyone can name why.

The agent runs unattended and the output needs a human. Splitting those two things properly is what keeps it alive past month two.

Who owns the run. Whoever owns positioning or marketing owns the context file and the material-change definitions. If that's you, it's a two-hour setup with a quarterly revisit rather than an ongoing task.

Who clears Answers. Whoever owns the surface being changed. A pricing page Answer goes to marketing, a call script Answer goes to whoever runs sales calls, a proposal template Answer goes to whoever writes proposals. Answers that come back to you every week are a sign the context file's constraints section is wrong.

Who decides Moves. You. Every time. A Move item is on your agenda within a week of appearing, and the decision gets recorded next to the gap that prompted it, which over a year gives you something more valuable than any single report: a record of every competitive decision you made and what the market looked like when you made it.

The review that keeps it honest. Once a quarter, read the last twelve reports and count the Answers that were cleared against the ones that expired. A high expiry rate means the agent's material threshold is set too low and it's promoting Noise. The fix belongs in the context file, not in the prompt.

You should have an agent like this running on your behalf, so the deals you are already in stay winnable and your own claims stay current against the market without you booking a quarterly review to do it.

You can build it from the prompts above. If you'd rather skip the setup work, Cortex subscribers get the full Competitor Watch Agent Build Pack as a download:

This is only available to paying Cortex subscribers.

Cortex is the premium monthly membership version of Bionic Business, helping entrepreneurs and business owners grow and scale with AI and Agents.

In addition to 3-4 full, unrestricted regular issues per month, you also get:

So, if you’d like to receive full issues and the monthly special Signals and Circuits issues, then you should sign up for Cortex, which is $50/month (and you have full archives of previous regular issues for as long as you’re a paying member).

One note on Signals and Circuits: You only get these special issues for the months you’re subscribed—which means if you skip a ~30 day period, about a month, you won’t get them for that month. There are no back-issues for past months, so now is your chance to lock-in these upcoming issues while you can.

After you’ve upgraded to Cortex, log in to the website and come back to this issue, and the Agent Build Pack will be unlocked below.

Ready? Get the Agent Build Pack, full regular issues, and the upcoming Signals and Circuits issues:

Cortex is a monthly premium version of Bionic Business, helping entrepreneurs grow and scale with AI and Agents. You can cancel at any time. But there are no refunds, for any reason. All sales are final.

The agency owner rebuilt his around the claim sheet in about an hour, and the first run told him two things his old reports never could.

Two of his five competitors had moved their entry tier below his in the same six-week window. And his own pricing page still described an onboarding process he'd stopped running in the spring.

Both facts had been sitting on public pages for months, and the monitoring he already had was pointed at them the whole time.

What changed is that the sheet put his own row next to theirs. A difference he had been reading as their news came back to him as his own gap, with a date on it and a page to go and fix.

Talk soon,
Sam Woods
The Editor

.