20 emerging AI engineering concepts mapped: context folding, latent memory, parallel inference
MaryamMiradi · x · 2026-11-15
Researcher Maryam Miradi curates 20 emerging AI engineering concepts for 2026 with GitHub repos, grouped into five areas: context compression & adaptation (latent context compression, context folding, agentic context engineering), adaptive memory & experience (latent memory, test-time adaptive memory, learned memory management), reasoning beyond written chains (recursive language models, continuous thought, soft thinking, recurrent-depth reasoning), agent learning & self-improvement (reflective evolutionary optimization, harnessed agentic RL, RLVR, self-play), and parallel generation & faster inference (diffusion LMs, speculative decoding, parallel drafting). She notes some are still research-stage and should be chosen by bottleneck.
More from coding & agent
- Top engineers stopped reading code as agents push every industry up an abstraction level — RaphaelDabadie · 2026-11-15
- Obsidian ships seven Skills that let agents fully operate your vault — MxMnr · 2026-11-15
- Clairvoyance launches free desktop app to orchestrate Copilot, Claude Code, Codex, Grok and local AI — draginol · 2026-11-15
- Vibecoding beats general-purpose software: dev builds his own timelapse video editor — mishig25 · 2026-11-15
- After 2 Years of Building AI Agents, He Landed on Go + Temporal — littlebobbyt · 2026-11-15
- E2B sandbox pattern in Alchemy: build a template snapshot at deploy, fork at runtime — samgoodwin89 · 2026-11-15