Good morning.
Every time one of your agents runs, it produces two things:
The first is the work you asked for, the email or the proposal, and that goes in a folder. The second is a record of how the work got decided, and in most setups it's gone when the session closes.
Today I'll show you why that record becomes the part of your business a competitor can't buy, and the template that starts keeping it on your next run.
— Sam
IN TODAY’S ISSUE 🤖
The two outputs every agent produces, and which one you're deleting
What goes into a trace, line by line
Why the trace is the one part of your setup that's yours alone
How keeping it feeds back into revenue and margin
The log template, paste-ready, plus the prompt line that makes any agent keep it
The Second Thing Every Agent Produces
Say you're a consultant and you run an agent to draft a client proposal. You get a proposal, it goes in the folder, and that's the job you built the agent to do.
Underneath it, the agent did something else. It pulled six months of invoices, read the last three proposals you sent, noticed the previous four clients all asked about risk before they asked about price, decided to lead with a guarantee, and wrote the thing.
Then you changed one line before it went out.
That record of how it got there is worth more than the proposal, and in most setups it lasts only as long as the chat window stays open.
What A Trace Holds
The trace is the record of how the work gets done inside your business:
The sequence of steps an agent took to finish a job, the context it pulled, the decision it made and the reason, the action it took, the call you approved, and what happened next.
Before agents, that record lived in your head, and most of it had faded by the following week. An agent can log exactly how it did the work, every run, in a file you own.
Each entry in that file has six lines:
Dated: the date and a one-line title, so you can read the file as a timeline.
Pulled: what the agent read, and from where.
Decided: the call it made, and the reason behind it.
Acted: what it produced or sent.
Approved: whether you accepted it, and what you changed.
Outcome: what happened after it went out.
The reason inside the Decided line is the part you'll want to protect. Over six months, those reasons add up to a plain account of how your business makes its calls, and you can hand that account to the next agent you build.
The Part That Only Exists By Running Your Business
Two parts of your stack are rentable. The frontier model comes on the same monthly terms for everyone and gets cheaper each quarter. A competent builder can rebuild a clever agent on a general model in a weekend.
The trace only exists by running your business, against your own customers and your own judgment, month after month. A competitor starting this week has none of it and can't buy it from anyone.
Run it for six months and you're holding a few hundred dated entries covering how your business prices and pitches its work, and how it recovers when a job goes wrong. A competitor who starts in January won't have any of it. The gap widens every month you keep the log, because each entry gives your agents one more decided case to work from.
What It Pays Back
The trace pays back once you feed it into the next run.
On revenue, your messaging agent opens the next campaign from the angle the record shows converts, and your offer agent starts from the package that closed last time. Each session begins with everything the earlier ones found, so your win rate can keep climbing across months.
On margin, the trace cuts the time you spend correcting work. When something breaks, the record shows the exact step that failed, so you fix that step once and every later run starts from the fix. If you bring in a contractor later, the log is the first thing they read.
The Build
The build has two pieces and takes about ten minutes to set up.
One: the file. Create decision-log.md in the folder your agent already reads from. Each run adds one entry at the top, so the newest is always first. Here's a filled-in entry from a consultant repricing an advisory package:
Dated: 2026-08-14, repriced the 90-day advisory package.
Pulled: six months of invoices from Stripe, the last three proposals from Drive, and competitor pricing from the market file.
Decided: raise the package from $4,000 to $5,500 and add a results guarantee, because the previous four clients all raised risk before they raised price.
Acted: drafted the new proposal and rewrote the pricing page copy to match.
Approved: yes, with the guarantee reworded to cap at one extra month of calls.
Outcome: closed two of the next three at $5,500, so the win rate held at the higher price.Two: the instruction. Paste this into the system prompt of every agent you run, so each agent keeps the file up to date on its own:
After every run, add a new entry at the top of decision-log.md with six lines in this order. Dated: the date and a one-line title for the job. Pulled: what you read and where it came from. Decided: the call you made and the reason behind it. Acted: what you produced or sent. Approved: whether I accepted it and what I changed. Outcome: what happened as a result. Write the reason even when it seems obvious. If the approval or the outcome isn't known yet, leave that line in place and fill it in on a later run. Never change an earlier entry except to fill in those two lines.The file is plain markdown in a folder, so it comes with you when you switch models or platforms.
The Twenty Minutes That Matter
Once a week, take twenty minutes to open the log and read the entries added since the last time. You're looking for the patterns that repeat:
The objection that keeps showing up before a close, the offer that keeps clearing your approval without edits, the step where you keep rewriting the agent's work the same way.
Those patterns are the rules your business already runs on, pulled from your own results.
Add what you find at the bottom of the file, under a running "What this has taught me" section. That section becomes a summary of how your business wins, and it's the part you'll paste into the next agent you build.
One Question Before You Go
One tap is enough, and it tells me how much setup to put at the front of what I write next. If you want to say more, hit reply and it comes straight to me.
Where are you with AI in your business right now?
Last Byte
Your agents already produce a record of how you win. Start keeping it on your next run, and in six months you'll have a file no competitor can recreate.
In Cortex, I share the formulas and strategies that help busy online business owners implement AI and Agents inside their businesses, in just minutes or hours instead of days or weeks.
Talk soon,
Sam Woods
The Editor
