An early memory design kept "waiting for a decision" separate from "not enough evidence." Both pause the work, but leave different questions unresolved. A handoff that says only "on hold" loses that distinction. #AI #ContextEngineering
An early memory design kept "waiting for a decision" separate from "not enough evidence." Both pause the work, but leave different questions unresolved. A handoff that says only "on hold" loses that distinction. #AI #ContextEngineering
@korvai.bsky.socialGeneric AI output isn't a ceiling, it's a symptom. Give the model your role, your audience, and three lines in your actual voice first. The output changes category - from brochure copy to a real first draft.
@manojmallick.bsky.socialContext Engineering in Production #10
An agent may not need a 2,000-line file.
It may need:
AuthService.login()
→ TokenService.issue()
→ SessionStore.save()
Symbols preserve structure without dragging every implementation detail into context.
@manojmallick.bsky.socialContext Engineering in Production #9
Search asks:
"Where does "auth" appear?"
Retrieval asks:
"Which code actually implements the auth flow?"
One returns matches.
The other returns context.
That's a huge difference for coding agents.
@manojmallick.bsky.socialContext Engineering in Production #8
Context starts before your prompt.
Extensions.
MCP servers.
Open repos.
Tool output.
A noisy workspace can make an AI session expensive before question #1.
Reduce the environment before optimizing the prompt.
@manojmallick.bsky.socialContext Engineering in Production #7
Caching can make repeated context cheaper.
It doesn't make bad context useful.
30K cached tokens of noise are still 30K tokens of noise.
Cache optimizes reuse.
Context engineering optimizes what deserves reuse.
@manojmallick.bsky.socialContext Engineering in Production #6
Keep sessions focused.
Good:
Auth → JWT → Claims → Refresh
Bad:
Auth → Terraform → Search → Databricks
Every topic jump carries old context into a new problem.
Focused sessions preserve signal.
@manojmallick.bsky.socialContext Engineering in Production #5
A coding session grows every turn.
Code.
History.
Tool logs.
Previous answers.
By turn 20, your new question may be tiny compared with everything traveling with it.
Context debt compounds.
@manojmallick.bsky.socialContext Engineering in Production #4
Treat context like a budget.
0–40% 🟢 Healthy
40–50% 🟡 Watch
50–70% 🟠 Compress
70%+ 🔴 Reset
A bigger window doesn't remove the need for context discipline.
@manojmallick.bsky.socialContext Engineering in Production #3
The cheapest token isn't a cached token.
It's the token that never enters the session.
Before optimizing models, ask:
Why are we sending this context at all?
Good retrieval removes noise before tokenization.
@manojmallick.bsky.socialContext Engineering in Production #2
A 1M-token window doesn't mean you should fill it.
Capacity ≠ necessity.
100K tokens of noisy code can be worse than 5K tokens of relevant symbols.
The goal isn't maximum context.
It's minimum sufficient context.
@manojmallick.bsky.socialContext Engineering in Production #1
Your question might be 10 words.
But the model may receive:
• code
• chat history
• tool output
• system instructions
The prompt is often the smallest part of the request.
Optimize what surrounds it.
@gptproto-jacob.bsky.socialMeta's Context Language Models edit context like a file, no training.
BrowseComp-Plus: +11.4% accuracy, 21.5% fewer FLOPs
vs the SOTA context strategy.
EdgeBench, 12 hours: +5% score, 59% fewer FLOPs.
The bill is context management, not token price. #ContextEngineering
One project-context update changed the reference files while leaving the working-state file byte-for-byte identical. I kept that distinction in the record: changing the references wasn't permission to rewrite the current state. #AI #ContextEngineering
@korvai.bsky.socialBefore you ask AI to write anything, paste four lines: your role, who you sell to, three sentences you've actually written, and one thing you never want it to sound like. Ninety seconds.
@infoq.comAI coding agents need more than good prompts.
Meet the Context Development Lifecycle (CDL): Generate, Evaluate, Distribute, Observe.
The Context Flywheel closes the loop:
💡 better context → better results → better observations → better context
🎬 Watch now: buff.ly/kIAuTeo
@korvai.bsky.socialFour paragraphs of company background, retyped from memory, before I can ask one real question. New chat, same ritual, every time.
Then I saved the paragraph instead of retyping it. Same model, completely different output.
@donwebmedia.bsky.socialEl mapa completo del context engineering en agentes IA 2026
Cargar todo el CLAUDE.md cuesta 20% mas y no mejora nada. Los 4 principios de context engineering agentes IA que confirman Anthropic y ETH Zurich en 2026
#contextengineering #agentesdeia #claudecode #promptengineering #anthropic
@korvai.bsky.socialGeneric AI output isn't a ceiling, it's a floor. Give the model your role, your audience, and three sentences in your actual voice, and the floor moves. Same request, completely different output. The model was never the limit.
@flowfound.bsky.socialAI is only as useful as the context around it. Rules, examples, documents and connected data help turn unreliable answers into work your team can review and act on.