GitHub decodes the new AI dev lingo: loop engineering, harnesses, squads and hill climbing
GitHub Blog AI/ML · rss · 2026-09-03
A companion guide to the latest GitHub Podcast episode breaks down the AI development vocabulary developers are encountering right now:
- Loop engineering: designing repeatable systems around agents instead of one-shot prompts — scheduled loops that fetch tasks, run the agent, validate output, and escalate failures, essentially AI-native cron jobs enhanced with skills, observability, routing and checkpoints.
- Ralph loops: the brute-force variant, where an agent keeps working from a spec until done. Useful for splitting large tasks into plan-act-check cycles, but token- and context-hungry.
- Squads and fleets: squads are role-specialized multi-agent groups mirroring real teams (planner, vet, implementer, tester, reviewer); fleets run agents in parallel. The core idea is parallelization plus specialization.
- Harnesses: everything around the model that makes it useful — tools, permissions, memory, context, orchestration. GitHub Copilot is the canonical example; "harness engineering" is improving that surrounding system.
- Hill climbing: iterating on agents and harnesses using evals to measure whether outputs actually improve.
- Forward deployed engineer: an existing customer-facing role rebranded with an AI focus, helping teams integrate AI tools and agents.
- Model openness spectrum: closed models (API-only), open weights (downloadable weights, training data may be withheld), and open source models (everything inspectable and modifiable).
The authors' takeaway: buzzwords fade, but the underlying practices — reliable repeatable workflows, validation, human oversight, and continuous improvement — are what matter.
More from coding & agent
- Dev hooks a busy-light to Gumloop to watch his agents work in real time — aronkor · 2026-09-03
- More MCP tools backfired: knowledge graphs fixed shallow agent answers in a stock research MCP — SnowSilent7695 · 2026-09-03
- Eve pitches itself as 'Next.js for agents': one-folder agent framework, durable by default — cramforce · 2026-09-03
- Claude logs into your Wi-Fi and fixes it for you — user builds an agent message board — scaling01 · 2026-09-03
- Muse Spark 1.3 ships with an underrated result, one-line curl install for Muse Code — alexandr_wang · 2026-09-03
- xState author opens experimental @xstate/effect PR bridging Effect 4 with XState v6 — DavidKPiano · 2026-09-03