MatIntellect

While the AI is working, you drop out of your flow state and your attention scatters, and that drives burnout up

Let me tell you how it was for me. I went through these stages:

• FROM a huge number of very different tasks/projects that lead to burnout (I had up to 6 projects at once)
• TO the optimal number I can work on at the same time, stay in flow and not burn out (now it's 1-2 projects at once)

I made a decision for myself: not to chase doing a lot and fast, but to do one project (two max) as well as I possibly can from start to finish and only then take the next one. That's how the burnout went away, and it keeps me in flow all the time 🔥

These days the work is mostly split into stages:

Architecture - 70-80%

1. Idea - no screen at all, just pure thinking in my head about how it should be built in the final version as I imagine it
2. The deepest research possible - launching a swarm of agents on the questions: how to solve this problem, are there competitors, are there ready-made solutions, what ready tools and combos exist, what has to be built from scratch
3. Prototyping and working through the idea - what makes a good MVP and what makes a finished product
4. Planning the implementation, security and testing - how everything will be done
5. Audit - I have an automated process of 8 reviewers with independent checks, with 3 models inside: Claude + Codex + GLM
6. Weak spots - meticulously working through every flaw the audit found

Development - 20-30%

7. Development - through GSD: research again, answering the researchers' questions, planning and so on through the GSD process
8. Testing - after every stage, fixing bugs right away
9. Live mode - the project running for real with built-in analytics at key points
10. Debugging + evolution - improving the product, building new features

What this order looks like on real projects

Take a chatbot for a clinic. Architecture took most of the time: what questions clients ask, where the line is between what the bot answers on its own and what it hands to the owner, where it gets prices from. In the end the bot covers about 114 price items and 48 services and answers only from the clinic's documents. What isn't in the documents it doesn't invent, it hands it to the owner as a card. I tested it not on made-up examples but on real questions from the clinic's inbox. Development like this takes about 2–3 weeks, and almost all of that time goes into thinking, not code

Another example is video subtitles. I built my own assistant in a day, because I knew the task inside out. For a client the timeline is from 3 days: first going through their task and testing on their videos, and only then building. Same principle: the better the task is understood, the shorter the development

How not to burn out when agents work on their own

I now have 8 Claude Code agents working every day, and the temptation to launch them on ten tasks at once is huge. But I keep the rule from this note: one or two projects, the rest waits in line. Agents can work in parallel, my attention can't, and it becomes the bottleneck

The second rule is not to keep everything in my head. Whatever can go to a schedule, I hand over: my personal agent Jes handles 10 scheduled jobs and reminds me of things I would forget. My head is freed up for architecture, not for a to-do list. And anything risky, costly or public the agent still does only after my yes

Bottom line - in 95% of cases I get a finished product on the first try and don't burn out 🎯 Working in GSD frees up more time to do something else in parallel, but I still keep an eye on the screen and sometimes catch bugs on the fly if development heads the wrong way. I never leave everything fully on autopilot - the cost of a mistake is way too high and it's a shame to lose the time ⏳

Share:

No comments yet

Leave a Comment

Fields marked with an asterisk (*) are required