My CI doesn't run in CI anymore
~100 PRs/day land on main. The setup:
- agents run the full test suite on their own “sandbox” before asking to merge
- a merge train batches whatever's ready and proves the batch once, ~12 min, on one hetzner box. Red batch → it bisects and kicks out the culprit back into the agent who shipped it and wakes it up
- releases go out continuously
And release management is an agent too. Every 20 min it checks the pipeline, sends failed batches back to the agent that wrote the PR, and fixes the pipeline itself
[fable 5.2 high] … “is the top of the Pareto curve. Often, Fable-class models are incredibly powerful but prohibitively expensive - this was not the case here.”
Replying to @ValsAI
The model is quite cost effective on “high” reasoning level, at only $17.94 / test, this makes it cheaper than both Opus 5.5 and GPT-6 Astra.
It is the top of the Pareto curve. Often, Fable-class models are incredibly powerful but prohibitively expensive - this was not the case here.
This is partially driven by Anthropic dropping the cost by 20% per-token: from $10 / $50 to $8 / $40 per million tokens.
3
468
multiplayer ADE level up: chatting with your team in your ADE
now available in Polaris @plrs_sh
cli ≫ mcp
you can now give your local agents polaris cli to handoff work to the cloud
try it out here -> plrs.sh/
what to show in the sidebar when an agent replies but its subagents keep running?
my fave is C
2
101
live machine resizing is now available in Polaris
Polaris lets you run coding agents in the cloud using your existing subscriptions
you can now resize their machines on the fly
is this what you get from an "autopilot for software engineering"?
i wouldn't expect the insides of a "high octane Ferrari" to look like this
pricing.html
> /* ========================================= Navigation (matches the main page nav exactly) ========================================= */
Launching @vorflux : The autopilot for software engineering. I was prev co-founder / CTO of @Rippling ($10B) and #1 coder in India. Vorflux is my high octane Ferrari.
Every AI coding tool still makes you fly the plane. That's the copilot model: you stay in the seat, approving every turn. The models quietly got good enough to fly the whole route, but the tools never caught up.
So we built the autopilot.
@vorflux raised a $15M seed by @ycombinator @peakxvpartners @alliancedao @parkerconrad @jake_zeller @balajis @nivi @metakovan @lmrankhan @nikitabase @0xrwu @ayushjaiswal @mattshumer_ @eshamanideep @sreeramkannan @dvcoolster @nusimow @TeddySolomon11 @ashtoncofer @rvivek etc
Drop your biggest engineering bottleneck below. I'll reply with how I'd attack it with Vorflux, and hand you $200 in free credits to bang out your backlog.
Our full thesis 🧵👇
300
> If this gets blocked, I owe you a reset.
this industry is brutal
If you aren't yet bold enough to install the Codex app, you can stay in the presence of your orange crab and point it at GPT 5.6 Sol. Takes 5 minutes. Kudos to Theo for explaining one of the ways to get this done.
Step 1: Install CLIProxyAPI
Step 2: Connect
Step 3: Define following alias and enjoy claudex
```
alias claudex='CLAUDE_CODE_SUBAGENT_MODEL=gpt-5.6-sol \
CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1 \
CLAUDE_CODE_MAX_TOOL_USE_CONCURRENCY=3 \
ENABLE_TOOL_SEARCH=false \
claude --model gpt-5.6-sol'
```
If this gets blocked, I owe you a reset.
2
264

















