First mover advantage works both ways.
- I got used to Claude code in cli so it is always awkward to use any other cli agent
- Recently, I and probably many others are getting used to Codex Desktop app so it is going to be awkward to switch to any other desktop coding apps.
It is so refreshing to read some engineering blogs published before 2022. Caring about engineering craft, no FOMO on AI and there is actual style to the writing. That world was so near yet so far away.
# of PRs per eng per week:
- Meta, 10
- Anthropic average, 30
- Claude code team or Replit, 100
It challenges me to rethink what does velocity mean and how to restructure the engineering team and process to facilitate the velocity target your team has decided
My team and I used to enjoy talking about, “What will you do when AGI is here?”
Turns out the better question was, “What will you do when the layoffs are here?”
Silly us.
A one-line change of basically A or B or C that AI keeps trying to make it a helper with 100 lines to test
I have to reach for VSCode/vim to manually type the changes in
Is this a good use of eng time or should we just give up and try to obey the AI verbose coding style?
Give this prompt a try:
From this point on, use a subagent to track the progress of this conversation and update ~/report.html
It is interesting that you basically get a journal for free
Evolution of my frontend questions:
- tell me about event delegation
- differences between state and props
- redux vs whatever is trendy that year
- rest vs graphql
- pros and cons of hooks
… many years later …
- what libraries are used in this app, did you choose them or did the AI choose it for you?
How do you even do frontend coding interviews those days? I’d love to hear tips
What do you say to your agent when it is time to execute:
- do it
- LFG
- Approved. Please go ahead. Thank you.
- Execute the plan. Do not make any mistake. Do not stop until you meet all acceptance criteria. If you failed, you will be fined $10M.
Did you know React’s launch was met with a lot of skepticism around why “rethink established best practices”?
Ironically, it is more relevant than ever to question established patterns:
- why do we staff multiple engineers per project. Kilo Code does one eng per project.
- why do we need code reviews
- why do we do quarterly planning. Claude Code core team meet weekly, set a theme for the week and then go.
- why do we have a team size of 10 per EM. Meta’s 1 to 50 org.
It is unclear what the new world will be but the first step is be willing to question and rethink well established best practices.
I must have committed unimaginable sins in a past life.
My punishment is modifying yet another implementation of the agent loop every time I start a new project.