the interactive mode is unusable with recent updates. the amount of dumb shit questions it produces is insane.
claude -p with opus 4.8 is a completely different world. the model feels a lot better, faster than earlier models, completes complex tasks in 10 minutes fully autonomous.
it's so dumb that they'll punish claude -p users soon with the extra special quota that burns likely in 5 minutes.
@burkov The problem is that corporates have started to limit AI spending. Devs are looking for cheaper models to get more tokens with the new budget.
Well Anthropic might still grow ofc but not at the pace they’ve grown in the past 12 months.
Commodization happens faster this time.
We keep hearing about 10x or 100x productivity gains in engineering and knowledge work.
But outside the model labs, I haven’t seen the corresponding 10-100x revenue growth across the market or increase in quality.
So where is the productivity going?
@thekitze Uhh it’s the opposite… agents are way better with statically typed languages than dynamic/untyped. Not cool to hallucinate variable / method names.
If you become exceptional at managing agents, but are also exceptional in your understanding of the fundamentals, you will be unstoppable.
We all prefer to work with masters of their craft. What’s new: you can’t afford to miss out on the amplification agents have on your output
I run Claude Code mostly unattended because it's very time-efficient. I plan several rounds, then let the agent do the work autonomously.
I get 80-100% done in one go. Having several parallel runs against real dev setups is what makes this a killer approach. I believe autonomous coding agents is the future, liked it or not.
Interactive Claude Code use has dropped to bare minimum in my workflow. After the autonomous agent finishes, I start digging the results usually in interactive mode, but this is maybe 10-20% of my workflow.
Also after the Opus 4.7 launch CC interactive mode has been more frustrating than before. Something has happened. Many are definitely not happy with it.
Like Conductor and others, I use claude -p to make Claude work automously. I developed a platform called https://t.co/XD0lJQyjJK that I use for both side projects and work work, also develop the platform itself with it.
Trimo runs agents in Docker containers without the limitations of git worktrees or the hassle that goes into starting and stopping parallel dev envs. Just type a prompt and go work on some other task.
Trimo still has some rough edges but you can try it if you've got GH CLI, Docker Desktop and Claude Code setup on your machine.
Getting started guide here: https://t.co/kZC7J0s5Pf
Drop me a DM if you need help with it.
Codex support coming sooner or later to Trimo.
codex is the best AI coding product and we want to make it easy to try.
for the next 30 days, we are giving companies that want to try switching over two months of free codex usage.
@ctatedev Every existing framework has been designed for humans. What would an agent-first framework look like? Maybe a very thin library and a bunch of MD files?
ArrowJS might be worth checking out: https://t.co/1aCYEkseth