Engineering Leader | Full-Stack Architecture & Team Growth | AI Platforms | 15+ yrs in tech | Rust | TS | Sui | I turn chaos into architecture.
liveness @iproov
@FactoryAI This is huge. Routing is honestly the only way to keep margins sane once you start scaling agents. 25% back in the pocket without losing quality is a massive win. 🛠️🦾
@garrytan honestly the "moat" strategy feels like a losing battle for labs long term. once the gap between models narrows, the winner is whoever builds the cleanest dev experience. 2027 is gonna be wild.
@marclou the first dollar is always the hardest but feels the best. stripe notifications hitting your phone is a better high than coffee lol. keep pushing 📈
@openclaw@Microsoft huge win for the dev community. seeing open source tools get first-class support on windows like this makes life so much easier for enterprise teams. congrats on the partnership! 🦀💻
@amytam01@SpaceX@xai huge congrats on the move! 🚀 being at the intersection of spacex and xai sounds like a wild ride. can't wait to see what you guys ship next. ad astra! ✨
AI agents need a substrate, not another memory SaaS. AllSource = high-perf event store (Rust, 12μs, MIT) + an ecosystem on top: Prime for hybrid recall, mammoth for coding agents, chronis for tasks. Agents self-provision in one POST + pay per query via x402. https://t.co/Qofxf35AAj
@SebJohnsonUK@runwayml@KanishkaNarayan london is absolutely cooking lately. having runway's world models team right there is a massive win for the local dev scene. the talent density is getting wild 🇬🇧🚀
A skill is just a folder with a SKILL.md. Install one: cp -r skills/stack-agnostic/rust-perf ~/.claude/skills/ — twelve skills, curated by exclusion. Repo + the full rejection criteria: https://t.co/AxNrmgZ9Mm Tell me which one you'd cut. ⭐
I shipped a curated set of Claude Code skills. Not a mega-pack. Twelve. Each one does something a stateless prompt fundamentally *can't* — carry memory across sessions, enforce a proof artifact, or wrap a real binary. The value is the no's. 🧵
The stack-agnostic six need none of my toolchain. e.g. verify-db-change proves a migration against real Postgres with a PR-ready block — schema per tenant, behavioural SQL inside BEGIN…ROLLBACK, reversibility against a throwaway clone. No more vibes-only "looks fine."
@fayazara Voice agents on the edge is the topology most teams underestimate. The latency budget for natural conversation is brutal, and putting the runtime next to the user is one of the few ways to actually hit it. Curious how this handles partial-transcript interruptions in practice.
@astral_sh Big news. The signal here isn't just an acquisition, it's that the AI infra layer is consolidating around teams who know how to build correct systems fast. Putting uv and ruff's DNA inside Codex changes what's plausible for the Python+AI stack. Congrats to the Astral team.
The amnesia framing matches what I see on my teams. The scheduled job approach is clever, but the harder question is what to keep vs what to forget. Most teams I've watched try this either keep too much (slow, expensive) or too little (the agent loses the throughline). The compression policy is the actual product.