Most automation never finds out how it went. So it never gets better. Here's how to build the feedback loop in.
Ashkaan Hassan · We Solve Problems · EO Los Angeles
You want every lead answered in five minutes. You can't staff that.
You want the numbers reviewed every Monday. Someone has to build it.
You want the follow-up done. Someone has to remember.
You want to enter a new market. Who's going to run it?
It was never the ambition. It was the capacity.
The good people were always expensive, already busy, and eventually gone. Now competent work is something you buy by the unit. It doesn't sleep, doesn't quit, and gets cheaper every quarter.
The question stopped being
“Who do I hire to do this?”
“What should own this?”
Almost nobody has rebuilt around that yet. The ones who win won't be the ones with the best model. They'll be the ones whose functions get better every week without them.
I run a small IT firm, and this system runs alongside me every day. Everything I'm about to show you is in production, wired into the real business.
of automations in production
tool I use, wired in
running unattended, on a schedule
vendor lock-in, all plain text
It runs my company and my house. None of this is a demo.
Every one is a yes-or-no you can answer about your own setup right now.
Does it ever find out how it went?
It fires, produces something, and never learns the result. This is almost everything people build. It plateaus, then gets quietly abandoned.
It sees what happened and writes the lesson down. You ignore it for a month and come back to something better than you left.
A function without a loop decays. A function with one compounds. That's the whole talk.
A loop needs somewhere to write the lesson down. These six pieces are that somewhere.
Your AI behaves the same way every time — your way.
Think of it as an employee handbook. It covers how you communicate, what tools to use, what rules to follow, and what's off limits.
You write it once. Every conversation after that follows your rules.
Output matches your tone — direct, concise, technically deep. No more editing AI drafts to sound like you.
The AI delegates research, bulk edits, and scheduled tasks to specialist agents. It manages them like you'd manage a team.
Every session follows the same lifecycle: load context, do the work, log what happened, save everything. Automatically.
The AI never sends emails without approval, never exposes credentials, never takes an action you haven't authorized.
Different AI agents can get different instruction files — a research agent gets different rules than your primary agent.
The AI always has exactly the right information for what you're doing — without you asking for it.
Writing an email? It already loaded your email templates. Talking about a client? It already pulled their profile.
It never loads everything at once. It stays fast and focused by only pulling what the moment requires.
Everything comes out in your voice — on the first draft, every time.
Your email style, code conventions, presentation format, communication tone — captured once, applied forever.
No more editing AI output to sound like you. It already does.
The AI knows your world — your people, your numbers, your goals. It never asks "what do you do?"
Mention a person — the AI knows them. Discuss finances — it has your budget. Context shows up automatically.
Work happens while you sleep. Reports generate, systems get monitored, content gets drafted — all on a schedule.
The AI understands each automation and can modify, debug, and extend them on its own.
I have dozens of automations running right now — from news digests to financial reports to health tracking.
Your AI talks to all your tools — pulling data, pushing updates, and orchestrating across systems you already use.
Your AI becomes the nervous system tying all your tools together.
"What did we decide about the pricing model?" Six months later, the AI knows.
Every initiative is tracked as a project with status and next steps. Every session ends with a journal entry logging decisions and changes.
The AI's memory gets better over time — not worse.
Each initiative gets a folder with status, description, and next steps. The AI updates these as work progresses.
Daily session logs capture every decision, change, and lesson. Searchable forever. Every session compounds on the last.
Consistent behavior, every time
Everything in your voice
It knows your world
Work happens while you sleep
All your tools, connected
It remembers everything
All plain text. All in one folder. All version-controlled.
No database. No proprietary format. No vendor lock-in.
One orchestrator, multiple specialists
Your primary AI doesn't do everything itself. It delegates to specialists — just like you manage a team.
Each specialist has its own instruction file. The primary agent only receives a summary back — keeping it fast.
Every session builds on every session before it. That's the compounding effect.
What turns an automation into a function that trains itself
Most automations have zero of them. Miss any one and the loop is open.
Did it find out how it went?
The correction you gave. The deal that closed. The alert that fired.
Did the lesson get written down?
A rule in a file. Not a chat message that scrolls away and is gone by morning.
Does the next run read it?
The rule loads before the work starts. So the mistake can't happen twice.
Signal without capture is a lesson you forget. Capture without recall is a file nobody reads. You need all three.
Not a diagram. This is a real loop from my system, running right now.
The AI asked to deploy something. I told it: a bug report is not permission to ship. Fix the tree, don't push until I say so.
That correction became a rule in a file, stamped with the date I gave it.
That rule now loads at the start of every session, before any work happens. I have never had to give that correction again.
Every rule in my system is a mistake that can't happen twice. The list only grows in one direction.
What this system produces in practice
Every Monday, 45 minutes pulling ticket data, calculating KPIs, typing numbers into a spreadsheet.
Ticket system → AI computes KPIs → pushes to scorecard. Every Monday at midnight. Zero human effort.
Every day at 7:00 AM, an email lands in my inbox:
I didn't ask for it. It just shows up.
Dozens of automations running 24/7. When something fails:
A monitoring system catches it immediately
AI diagnoses the root cause
Applies the fix automatically
Notifies me only if it can't self-resolve
Most failures are fixed before I even notice.
This isn't incremental improvement. This is a step change.
And it compounds — every automation you build frees time to build the next one.
How to build your first loop this week
The rungs aren't a staircase you have to climb in order. A loop can ride on a skill you still trigger by hand.
You run it yourself, it learns from each run, it gets better. No schedule, no infrastructure, no engineer required.
Which means the cheapest closed loop in the world is this: you as the sensor, a file as the memory.
Pick one thing you already ask AI to do. Close the loop on that one thing.
One empty text file where fixes live. This is the memory. Point your AI at it so it reads the file before it works.
Don't just fix it in the chat. Add the lesson to the file, in one line. That's capture. You've now closed the loop once.
In a month the file is your standards, written by your own corrections. That file is the function getting better on its own.
You ask, it answers, you do the work.
It runs a skill the same way every time.
Something other than you pulls the trigger.
Every run makes the next run better.
Your first loop takes an afternoon, and it doesn't need a single rung below it.
I open-sourced the framework behind everything in this talk. It's called Contextium.
Apache 2.0. Free forever. You own everything.
$ curl -sSL contextium.ai/install | bash One command. Five minutes. The floor is already built.
Open floor
Ashkaan Hassan
We Solve Problems · wesolve.tech · EO Los Angeles
contextium.ai — open source, free forever