A deliberate replay of an idempotent API endpoint that wasn't truly idempotent could generate $100M in duplicate charges before detection. #AIEngineering
@levelbrook.bsky.social
@levelbrook.bsky.socialA deliberate replay of an idempotent API endpoint that wasn't truly idempotent could generate $100M in duplicate charges before detection. #AIEngineering
@qconferences.comDan is speaking at QCon San Francisco, Nov 16-18.
@catchproof.bsky.socialBreakthrough in AI stability tracking 🚨
Barbara Roy released a new mechanism-layer operator to measure AI system drift before it compounds over long inference horizons.
Full paper here: The Stability Envelope™ doi.org/10.5281/zeno...
@jmm86.bsky.socialInfinite retries cost too much: they waste budget. Say goodbye to agents retrying indefinitely with a set limit for efficiency and effectiveness. #AIEngineering #EfficientAutomation
@yurekilab-en.bsky.socialLeft a "TODO: clean up later" comment in my code. My agent found it mid-bugfix and shipped an 80-line refactor I never asked for. Every comment is an instruction now, whether I meant it or not. TODOs get scope and boundaries, or they don't get written. #claudecode #ai #aiengineering
@qconferences.comJoey is speaking at QCon AI New York, Dec 15-16.
@yurekilab-en.bsky.socialI don't let my coding agent fix a bug until it writes a failing test that reproduces it. About a third of the time, that test fails somewhere I didn't expect — the bug wasn't where either of us thought it was. #claudecode #ai #testing #aiengineering
@ilgarz.bsky.socialTurning an AI prototype into something a business can rely on is a different problem entirely.
Only 50 seats available. Registration closes 1 October 2026.
RSVP: meetup.com/azdata-aicommunity/events/316287557
#DataAIDayBaku2026 #AIEngineering #ArtificialIntelligence #RAG
@58bits.comMy take on the frontier labs' public statements about a desired 'slowdown' in frontier model development? They're crack dealers. Don't believe it for a second ;-)
@ipolyzos.comRig brings familiar software engineering ideas to Rust AI: provider abstractions, typed tools, testing, RAG, streaming and local models.
Production AI needs more than prompts. It needs good software engineering.
@qconferences.comPlatform engineering now has another user: the agent.
Mark is speaking at QCon SF, Nov 16-18.
@brwngrldev.bsky.socialI've been digging into engineering metrics in teams that use AI heavily.
One weird finding: a PR with ~50 review comments looked risky until we opened it. Most of the comments came from AI.
I think we're going to need AI-aware engineering metrics, not just AI adoption metrics. 🔬
@qconferences.comHe’ll also cover MIPRO, GEPA, evaluation cost, and safe rollout patterns at QCon AI New York, Dec 15-16.
@heyymeraki.bsky.socialToday we are looking at the build for yesterday's introduction to ai engineering. If you read it and wanted to see what an LLM API call looks like in code, I go step by step so it is easy to follow. Read it here:
@yurekilab-en.bsky.socialStarted watching which files my coding agent reads first on each new task. It's never the module I'd call the core. It's the types file and one config — the real map of the project. The agent figured that out before I did. #claudecode #ai #aiengineering #buildinpublic
@geraldnguyensg.bsky.socialThe flashiest agent demo is not the durable advantage. As AI platforms formalize confirmations, traces, and resumable tasks, the real moat is the recovery layer that pauses bad runs before they become user harm. #Ai #AiEngineering #AppliedAi
@levelbrook.bsky.socialStreaming a conversational AI trainer, and teaching it to say I do not know mid-stream. Abstaining after you have already emitted tokens is its own problem. #LLM #AIEngineering
@yurekilab-en.bsky.socialMy coding agent now logs its decisions, not just its code. Each time it picks between two approaches, it leaves a three-line note: chose, rejected, why. Skimming 40 notes caught two quiet architecture drifts no diff review flagged. #claudecode #ai #aiengineering #buildinpublic
@gothartech.bsky.socialLet someone else worry about uptime?
#Gothar #AIEngineering #EdgeInference #PostSaas #SelfHostedAi
@yurekilab-en.bsky.socialI stopped interrupting my coding agent mid-task. Cutting it off at the first wrong turn left me half-applied edits. Now I let it finish, then read where it drifted — the detour usually points at a gap in my instructions, not the model. #claudecode #ai #aiengineering #buildinpublic