Day with AI: PRD for an Autonomous System, Broken Python Fix, and the Client Got an Implementation Plan.
Wednesday mix of everything — from autonomous system design through dev tool fixes to process analysis for a client. 6 sessions, zero commits, but lots of writing.

What I worked on
Wednesday. One of those days where you don't write a single line of production code but you're busy all day. Six sessions, lots of documents, a few fixes, and one legacy project from the past.
What I did
PRD for autonomous Work-OSBiggest task of the day. Wrote a complete product requirements document for making my personal operating system do things on its own — without being told what to do. Went through the entire current state, identified what's already automated and what isn't, and designed the target state. A meta-task: writing a document about how the system should govern itself.
Process analysis for a client — implementation planClient needed a concrete plan added to their existing analysis: what, when, and how much. Phased the implementation across 3-4 months so it wouldn't be a one-time financial hit. Each phase builds on the previous one and is independently functional — after phase one, the company is already using the system.
9 processes, from document processing through inventory to financial reporting. All in language a non-technical person can understand.
Dev tools fix — Python and observabilityLangfuse hook for tracking AI sessions stopped working. Root cause: Python 3.14 broke Pydantic V1. Fix: recreate virtual environment with a compatible version. Classic — half the work with AI is fixing infrastructure around AI.
Cold email pipelineMorning automated pipeline run. Scraping, data enrichment, draft generation. Runs in the background, I just check the results.
Design skills consolidationMerged several UX/UI tools into a single file. System maintenance — nothing sexy, but necessary.
Legacy Vue projectAn older project I manage. Vue app that needs to go as a static build to FTP. Found only net2ftp (a web FTP client) on the server, no production code. So a detective story — figuring out what's actually running there and how to deploy it.
Time: AI vs without AI
| Task | Without AI | With AI |
|---|
| PRD for autonomous system | 2-3 days | ~3h |
|---|---|---|
| Implementation plan (9 processes, phases) | 1-2 days | ~2h |
| Python/Langfuse fix | 2-3h | ~20min |
| Cold email pipeline (automated run) | — | ~10min check |
| Design skills consolidation | 2-3h | ~30min |
| Legacy Vue project analysis | 1-2h | ~30min |
| Total | ~25 hours | ~6.5h |
What I learned
- Writing a PRD for your own system is surreally meta — you're writing a document about how the system should write documents itself
- Python 3.14 breaks a lot of things — Pydantic V1 doesn't work, and Langfuse depends on it. Stick with 3.13
- Implementation plans for clients need to be phased — nobody wants to hear "160 hours upfront"
- Legacy projects without documentation are detective stories — but with AI, at least faster detective stories
Interested in the article?
Let's discuss what this kind of automation can do in your company.
Free consultation