I built a harness for the so-101 arm after getting frustrated with doing teleops and the arm not generalizing
the most surprising learning is that Kimi K3 is a lot better at this than Fable
Today we're launching Superposition. An AI headhunter better at finding talent than any human.
We made a short film to prove it. A fictional billionaire finally admits how he built his empire.
Spoiler: it was never about him. It was about the people he surrounded himself with.
Watch it. Then come find your perfect weirdos.
@joulee Thank you for this piece. Had similar thought today, from the lenses of someone who hasn't made it, and not really feeling envy / fomo this time.
It came from one idea that money doesn't really change you. I will still have the same flaws. Growth requires work and experience
We just shipped Webhooks for TabTabTab.
You can now trigger your agents from CI, internal tools, Slack, or anything that can make an HTTP request.
TabTabTab is always-on. When a webhook comes in, it can route the request to the right agent and let it use the full capabilities of its cloud environment โ understanding the context, taking action, running checks, and moving the task forward.
We use this a lot at TabTabTab, hope you also find it useful!
Cloud-first. Cloud-only. An Agent Development Environment we've been building in silence.
@HussainKarim10 gives you the preview:
โข Laptop dies, you keep shipping
โข On the move? Ship from your phone
โข Use your own Codex login
The machine is no longer the bottleneck.
coding agents can do a better job than me in keeping the github issues updated / organised as I work. I never had the attention to keep updating it in detail.
this is an example where I asked @TabTabTabAI to work on something end to end, in this case make sure bun is installed in the AMI. i told it to work in a new worktree, create PR, link it to the issue, and do full verification (create new AMI, provision new VM through local controlplane service, and smoke test the installation)
@gdb@saranormous There was a time when I was making a product and thought itโs useless without using (at the time) Sonnet 4. I thought there was no way to compete as Anthropic can subsidise their token cost. Turns out to be not true now, other models are just as capable
@simonlast Very much agree with all this. Sessions run longer now and more capable . I run my agents on the cloud with full VM setup, so it runs the full services and validation. I can steer and monitor progresses from my phone. Happy to set it up for anyone interested