You have been trying to get AI working for months. You have 12 tools, three half-finished workflows, and a growing sense that everyone else has figured this out and you have not.
The problem is not you. It is the approach.
Most solopreneurs add tools when they should be adding structure. The 90-day reset does not give you better tools. It gives you a working system — one function at a time, in a sequence that compounds.
The reset starts with a cut
Before you add anything, cut.
Open your subscriptions. Cancel every AI tool you have not actively used in the last 14 days. Keep a maximum of three. If you cannot decide which three, keep the one you have used most and cancel everything else.
This is not about finding the best tools. It is about having fewer decisions. Every tool you keep is a context switch, a login, a monthly charge, and a half-formed workflow competing for your attention. Cut the field. Then build.
Month 1: ship one thing per day
Pick one AI tool for your highest-volume task. If you write content, that is your content tool. If you handle client proposals, that is your drafting tool. One tool, one task.
Use it every day. Ship one output per day — a post, a response, a summary, a proposal section. It does not have to be perfect. It has to ship.
Do not evaluate alternatives. Do not optimize the prompt. Do not read about what other people are doing with their stacks. Just ship.
At the end of 30 days, you have 30 outputs and real data on what works. That data is worth more than any prompt guide you could have read instead.
Month 2: add the evaluation layer
You are shipping consistently. Now add the quality check.
Build a critic prompt for your highest-volume output. The critic prompt asks: does this output meet the standard? It checks for the things you care about — tone, completeness, accuracy, whatever matters for your function. Write it in plain language. It does not need to be sophisticated.
Run it on everything you produce. Measure the pass rate — how many outputs pass the critic without you manually editing the prompt. Refine the critic until 80% of outputs pass on the first run.
At 80%, you have a working evaluation layer. Below 80%, your base prompt needs work. The critic tells you where.
Month 3: document and expand
Write down what works: the prompt, the workflow, the evaluation criteria. One document. Plain text is fine.
This is your system. Not a framework borrowed from a newsletter. Your actual system, built from 60 days of your own outputs.
Now pick a second function and repeat months 1 and 2 for it. A different task, the same sequence: ship daily, then evaluate, then document.
By the end of month 3, you have two working AI functions and a documented system for each. That is the foundation.
What operational looks like
Operational means four things:
- You know which tool handles which function
- You have a prompt that works
- You have an evaluation step
- You ship consistently
It does not mean you have the best stack. It does not mean your prompts are elegant or your workflows are automated. It means the system runs without you having to think about whether it will run.
Most solopreneurs never reach operational because they keep optimizing before they have shipped enough to know what to optimize.
The compounding starts at operational
Overwhelmed produces no compounding. You are busy, but nothing accumulates.
Operational produces compounding from day one. Every output you ship is a data point. Every evaluation is a refinement. Every documented workflow is a reusable asset.
After 90 days of operational, your prompts are better than they were on day one because you have run them 90 times. Your evaluation criteria are sharper because you have applied them to real outputs. Your system is measurably better — not because you read about improvement, but because you shipped through it.
The gap between overwhelmed and operational is not talent or tools. It is 90 days of structured repetition.
We send a 90-day AI reset worksheet to Field Notes subscribers — the cut list, the month-by-month plan, and the evaluation criteria template. Get it at avakata.agency/contact.html.
What we did at Avakata
The Avakata engine started with one agent on one function. Month 1: ship. Month 2: evaluate. Month 3: document and expand.
18 months later: 160+ agents across engineering, design, data, marketing, sales, and support. The pattern is the same at any scale — one function, structured repetition, documented output.
We did not start with 160 agents. We started with one, and we made it operational before we added the second.
If you want to talk through what the reset looks like for your specific functions, book a discovery call. We will look at what you have, what to cut, and where to start.