An AI engineer that remembers every commit.
What does your coding agent know on day one?Same agent. Different brain.
Both run on the same AI models everyone already uses. The difference is what each one read before it started. Ask them both to fix the bug from the client's video.
Sure. Could you describe the bug? Which page is it on, and what are the steps to reproduce it? I'll start by reading the repo structure.
It's the double charge. At 0:41 in the video the client taps Pay twice on a slow phone. We fixed the same race in booking with an idempotency key, so same pattern here, test included. The PR is open.
The generic one asks questions. The Kortex one cites the video, the commit and the call.
What does an AI engineer need to read?
Code is the easy part. The why lives in calls, videos, tickets and old reviews. Plug in what your team already produces and watch what it can answer.
Kortex for business is a concept I'm building. The AI engineer on this page already runs in production for clients, reading exactly this kind of context.
The client's video is the ticket.
Clients don't write tickets. They hit record and talk. The agent transcribes every Loom, finds the moment that matters and matches it to the pull request that closes it.
Recorded on a phone, no ticket attached.
- 0:12So this is the checkout on my phone.
- 0:27The page is slow, so I tap Pay again.
- 0:41And now I'm charged twice. Every time.
- 1:05Customers are emailing us about it.
Matched: double charge on a second tap
fix: second tap on Pay charges once
- 3 files, 1 new test
- Tester: passed on a slow phone
- Reviewer: approved
The code gets merged. The lesson stays.
My agents work overnight as a fleet: parallel lanes write, an independent tester tries to break it, reviewers send it back. Without a Kortex, every lesson dies with the session. With one, tomorrow's lane starts smarter.
Every task that didn't teach the next one is half done. That is the whole idea behind Kortex.
Draft, Ask or Automate.
You choose per task, not per agent. Everything starts in Draft. As a kind of change proves itself, it moves up the dial. Trust grows one merge at a time.
Draft means it writes and you merge. Automate means you read the summary over coffee.
Payments, auth and migrations can live in Ask forever. Copy fixes and tests can go to Automate in a week. Your call, per kind of change.
fix: second tap on Pay charges once
This touches the payments table. Add the migration and go ahead?
Merged overnight: four pull requests. Tests green, reviews approved. Nothing waits on you.
It opens the pull request. You review and merge.
Not a demo. A shift.
Real numbers from real work. The client stays unnamed; the pull requests are very real.
- Client Loom videos transcribed and matched to the pull requests that close them.
- 975 learnings across 80 projects that my agents read before they start.
- 21 working demos built for prospects in a single day, by the same overnight fleet.
And the human behind the fleet: Top Rated Plus on Upwork, 100% Job Success, 11,303 hours.
What do people ask about AI coding agents?
What is an AI engineer agent?
An AI that writes, tests and merges code on your product. Mine run as an overnight fleet: parallel lanes write the code, an independent tester tries to break it, and reviewers check it before anything merges. You wake up to pull requests, not promises.
Can an AI coding agent work on an existing codebase?
Yes, that is the whole point. It reads the repo, the commit history, the tickets and what the client said on calls before it touches a line. For one client it merged 335 pull requests in 102 days, each one adding to what the next one knows.
How does it know what the client actually wants?
It reads what your team reads. Client Loom videos get transcribed and matched to the pull request that closes them, so the video becomes the ticket. Calls, tickets and past review notes sit in the Kortex it reads before it starts.
Does it merge code without me?
Only if you set it to. Every task sits on a dial: Draft (it opens the PR, you merge), Ask (it checks with you before anything risky, like payments or auth) or Automate (tests green, review approved, it merges). Everything starts in Draft.
Who checks the AI's code?
Not the agent that wrote it. An independent tester runs the change in a real browser and tries to break it, and separate reviewer agents read the diff. Self-checking is how agents ship bugs with a smile.
What is a Kortex?
One brain for a business. It sponges in calls, email, WhatsApp, Slack, tickets, docs and code, and every agent reads it before it works. Kortex for business is a concept I am building; the AI engineer on this page already runs in production for clients.
The long versions, from the blog.
Hire me or not: start collecting your data now. Everything.
A coding agent without a Kortex is a new hire every single morning.
Same brain, other hands.
Every agent below reads the same Kortex. Teach one, and the others know it by morning. See all the agents →