i believe this is as close to a software factory as you can get with plain claude code & cloud execution
the session is still unfinished after 24 hrs
a quick glance at session output reveals that:
- claude code's container ran out of free disk space thanks to stale worktrees
- auto-continue, if it exists, does not work (took 4 hrs to resume after limits reset)
- spends a lot of time running tests
- not ok to let it come up with its own sw dev practices
- anything claude does should be defined in advance: tasks, workflows, tools, deps
this does not scale at all. i wouldn't tell my team to start running long tasks like this with claude.
a complete app is too much asked.
it's better to stick to running a feature factory in claude instead.
i spent an hour writing a huge prompt in ios notes and pasted it to claude code around 8.30 this morning.
agents still working and it's 16.27.
i'm rebuilding a codebase from 2016-2018.
the task was to study the existing implementation, extract concepts / ideas / modules / features, and rebuild those from scratch using a modern stack.
cool to see big corps doing the same stuff that i've been cooking
sre agent monitors logs
-> detects problem
-> linear issue
-> study issue, fix it & pr with UI recording if needed
-> review (automated + human)
-> done
ok clairic is cool
rip structured output
Introducing Clairic, from Cloudflare. A System One extraction modeI, capable of extracting, computing, estimating, and forecasting data from unstructured inputs. cfl.re/4zIgfpH
i spent an hour writing a huge prompt in ios notes and pasted it to claude code around 8.30 this morning.
agents still working and it's 16.27.
i'm rebuilding a codebase from 2016-2018.
the task was to study the existing implementation, extract concepts / ideas / modules / features, and rebuild those from scratch using a modern stack.
The most humbling experience in a while: Google Ads.
The UI is a true final boss. The amount of layers is astounding. Wishing all the best to those who have to use it daily.
Gonna setup the MCP next...
revenue = product * marketing * demand
how much revenue do you get if your app has zero demand?
a product generates revenue because marketing captures demand
cercle archives is a goldmine for background music work / study etc.
usually instrumentals only, relaxed bpm, great views too
youtube.com/watch?v=cVFzblT5…
why personal software is not going to happen
Replying to @rywalker
it has to be super easy to go from intent to a working thing
i still buy bread even though i could buy a bread maker
d1 is cool but can get expensive if high traffic
asked claude to do some perf analysis and fix things with cf skills + cf cli
estimated cost was roughly 60 usd/mo
now zero after optimisation
If you can make users feel relieved, you're onto something.
The biggest wow effect comes from solving frustrating things.
Doesn't have to be groundbreaking or complex, just something that people consider tedious.
One of the best examples I know: managing a calendar with an AI agent.
Create a recurring meeting (think irregular schedule), invite people, set alerts, write description, remember to add Google Meet (yes you can forget that too).
All this takes a shitton of effort manually with a huge risk of failure.
Solve problems like that and you give users their time back. The faster you do this, the more likely they'll pay.
Think time-to-value, counting from e.g. signup / trial start.
Claude Code's auto-mode classifier has become conservative.
It waited two hours for my input in a background tab whether it is allowed to write a comment to a Linear ticket or not.
Used to work before but not anymore.
Thank you for wasting my time. 🫡
Literally impossible to find the billing settings in Bitbucket. Atlassian must have 0% churn. $TEAM











