June 11, 2026 | Read online

Welcome to The Ops Digest!

Each week, we drop no-BS insights and one AI prompt to cut wasted costs, tighten workflows, and eliminate manual grunt work.

The last few issues had you running AI workflows by hand. This one runs on its own.

Today: the competitor moves you keep finding out about too late, and a Claude Cowork agent that watches for them on a schedule and reports only what changed.

Your Competitor Changed Something Last Month. You Have No Idea What.

A competitor of yours cut their standard lead time from three weeks to five days back in March. They did not announce it. They updated the quote template and started telling buyers, one at a time. You found out in May, when a fifteen-year account asked you to match it and you could not, because you did not even know that was the new bar. The deal was gone before you knew there was a deal to lose.

This happens constantly. A competitor drops list price 8%. Adds same-day shipping on one product line. Changes their minimum order. Posts a job for a regional rep in your territory. None of it shows up anywhere you happen to look. It reaches you weeks later, secondhand, usually from a customer who is already halfway out the door.

You could assign someone to check competitor sites every Monday. They will do it for three weeks and then stop, because it is boring and nothing changes most weeks. So the job needs to belong to something that does not get bored. That is what this is.

The Job Nobody Keeps Doing

Every sales leader has at some point told someone to keep an eye on the competition. The person checks diligently for two or three weeks, finds nothing most days, and quietly lets it drop. Then three months later a competitor move blindsides the team, and everyone agrees, again, that someone really should be watching.

The work is not hard. It is boring and easy to skip, which is the exact profile of a job that should not belong to a person.

So hand it to something that does not get bored. Claude Cowork added scheduled tasks this year. Per Anthropic's own documentation, a scheduled task runs on a recurring cadence in the Claude desktop app, with access to your connected tools, and Anthropic lists tracking competitors as one of the intended uses. You describe the job once. It runs Monday morning without being asked.

The Memory Problem That Trips Everyone Up

Here is where most people set this up wrong, and it is worth getting right before you build anything. Each scheduled run starts a fresh session. The agent does not remember what it saw last week. So if you simply tell it to check your competitors every Monday, it will hand you a full summary of everything it finds, every single week. The same six pages, lightly reworded. You will read it twice and then start ignoring it, which is the exact failure you were trying to escape.

The fix is to make the agent compare, not just report. On each run it writes a short snapshot of what it found to a file. On the next run it reads that previous snapshot first, compares, and tells you only what is different. New price. Shorter lead time. A product page that did not exist last week. A job posting for a rep in your region. Everything unchanged gets one line, no change, and you move on.

That one design choice is the whole difference between an agent you keep and an agent you mute. A weekly wall of text trains you to ignore it. A short list of what actually moved, most weeks just a line or two, is something you will still be reading in two years. Build the comparison in from the start.

Build It: The Competitor Watch Agent

Setup runs about ten minutes. Open Cowork in the Claude desktop app, start a new task, and give it the job below. First, decide what it watches. You do not need everything. Four or five sources per competitor, the ones where real moves actually show up:

Pricing and lead-time pages. If a competitor publishes either, this is the highest-value thing to track. A list price cut or a shorter quoted lead time is a direct hit on your next quote.
Product and catalog pages. New SKUs, a new line, a product marked back in stock after a long gap. Changes here tell you where they are expanding.
The careers page. The most underrated source there is. A posting for a regional sales rep in your territory, or three warehouse roles at a new location, tells you their plans before their marketing does.
News, press, and LinkedIn. A new facility, funding, an acquisition, a senior hire. The announced moves, so you are not the last to hear them.
A short list of your own numbers. Give the agent your current standard lead time and list pricing as a reference, so it can tell you not just what a competitor changed, but whether they just passed you.

Then schedule it. Inside a Cowork task you can type /schedule and pick the cadence. Weekly on Monday morning is the right default for most teams. Anything more frequent is noise. Anything less and you are back to finding out late. The task prompt is what makes the comparison work, so write it once and leave it alone. Here is a version to start from:

You are my competitor watch agent. Run this every week. Step 1. Read the file competitor-snapshot.md if it exists. That is what you found last week. If it does not exist, this is the first run, so there is nothing to compare against yet. Step 2. Check these sources and note what you see right now: - [Competitor A pricing page URL] - [Competitor A lead-time or shipping page URL] - [Competitor A product or catalog page URL] - [Competitor A careers page URL] - [repeat the set for each competitor] Step 3. Compare what you see now against last week's snapshot. Write me a short report with ONLY what changed: - New or changed pricing - New or changed lead time or shipping terms - New products, new lines, or stock status changes - New job postings, especially sales roles or a new location - Anything that now beats our reference numbers below For everything unchanged, write one line: No change. Step 4. Save this week's findings to competitor-snapshot.md, overwriting the old file, so next week has something to compare against. Rules: - Report only what you can actually see on the pages. Do not guess and do not fill gaps from memory. If a page will not load, say so. Never invent a change. - Quote numbers exactly. Do not round. Our reference numbers: - Our standard lead time: [for example, 3 weeks] - Our list pricing: [for example, published on request]

How to Read What It Sends You

Most weeks it says nothing, and that is the point. A run that reports no changes is not a wasted run. It is the agent confirming the bar has not moved. The weeks it does flag something pay for all the quiet ones.
Separate a real move from a website edit. A competitor rewording their homepage is not a move. A lead time going from three weeks to five days is. Skim past the cosmetic changes and stop on anything that touches price, lead time, availability, or footprint.
A new rep posting in your territory is an early warning, not a today problem. It means they are about to push into your accounts a quarter or two out. That is time to shore up your at-risk relationships before the calls start, which beats reacting after.
When it flags a number that beats yours, that is the one to act on this week. Shorter lead time, lower price, broader stock. Get it in front of whoever owns the affected accounts before the next quote goes out, not at the next monthly review.
Spot check the first few reports against the live pages yourself. Open the competitor page and confirm the change it flagged is really there, and that a no-change week is really no change. If it missed an obvious update or invented one, fix the source list or the prompt now, before you start trusting it unread.

An Honest Take

The competitive intelligence industry will sell you a platform that tracks a hundred signal types across every competitor you have. I have watched companies buy those, use them hard for a month, and then let them rot, for the same reason the manual version rots. Too much input, no decision attached to it. The agent in this issue is deliberately small. Five sources, one question, only what changed. I think small and actually read beats comprehensive and ignored every time, and it is not close. The honest truth about competitive intel is that the hard part was never collecting it. It was getting a busy team to look at it every week without fail. An agent that hands you two lines on a Monday solves the part that actually breaks. The forty-tab dashboard solves the part that was never really the problem.

What to Watch Out For

It only runs while the app is open. Cowork scheduled tasks run on your desktop, and only while the computer is awake and Claude Desktop is open. If your machine was asleep Monday, the run happens the next time you open the app and you get a notification. Fine for a weekly competitor check. Do not rely on it for anything that has to fire at an exact minute.
Public pages only. The agent reads what a logged-out visitor can read. It cannot get behind a competitor's login, customer portal, or gated pricing. If their real pricing lives behind a quote request, the agent will not see it, and you should not pretend it can.
Invented changes. If a page fails to load, an agent can paper over the gap with a plausible guess. The prompt tells it to say so instead, but check the first month of reports against the live pages, so you trust the no-change weeks as much as the change weeks.
The snapshot file is the whole engine. If it gets deleted, or the agent cannot write to it, every run becomes a first run and the comparison breaks. You are back to weekly walls of text. If the reports suddenly get long and repetitive, check that the file is still there.
Do not over-monitor. Twelve competitors across fifteen pages each produces a report nobody finishes. Pick your three or four real competitors and the handful of pages that move. You can always add one later. Starting big is how this dies in week three.

Wire It Into the Monday Meeting

Run it once and you have a snapshot. The value shows up around week six, when it flags a competitor cutting their lead time and you hear about it the Monday it happened instead of the month after. But that only matters if the report reaches the people who can act on it.

So give the output a destination. Have the agent write the weekly report to a shared file, or point it at the channel your team already reads. The worst version of this is an agent quietly doing good work into a folder nobody opens. Put the two lines of what changed in front of the account owners before they send the next quote. If your plan includes scheduled runs, the task takes care of itself. If not, you open it once a week and run it by hand, which is still a minute of work for a standing competitive read.

And this is the shift worth noticing. The last several issues had you running AI workflows by hand, a project here, an audit there. This is the first one that runs without you. Once a competitor watch agent is sitting on your Monday, the obvious next question is what else on your plate is a boring weekly check that an agent should own instead. That list is usually longer than people expect.

The Bottom Line

Two-thirds of your deals are competitive, and competitors move quietly. The information is public. The reason you find out late is not secrecy. It is that watching for it every week is a job no busy person reliably keeps. So stop assigning it to a person.

Build the watch agent. Give it your three or four real competitors and the five pages that actually move. Make it compare against last week and report only what changed. Schedule it for Monday morning and route the result to the people who quote. Most weeks it will say nothing. The week it does not is the week you stop losing a deal you never saw coming.

Your competitors are moving faster. Shorter lead times. Cleaner quotes. Faster confirmations.

Some of that is strategy. But a lot of it is just that their team isn't still rekeying orders by hand.

If your people are spending their day on manual data entry, they're not watching the market. They're not catching the account that's drifting. They're not doing any of the things that actually keep customers from leaving.

Y Meadows automates order entry from intake through ERP.

Takes 20 minutes to see how it works.

👉 BOOK A DEMO

The deal was gone before you knew the rules changed