arindam

Generative AI solution engineer | Building Agents

[arindam @_arindam]

I think the final battle will always be with Gemini and Grok.

Claude will remain as a protected model and target enterprises with slow careful progress.

OpenAI may still lead on active user base since ChatGPT term became a synonym to any AI. Although their models may stay stagnant for sometime.

Meta may come-up with something. Time will tell.

[arindam @_arindam]

@arpit_bhayani Yes. CLI is like a gift for the Agents.

[Muratcan Koylan @koylanai]

Gradient descent for SKILL.md files sounds interesting, maybe a bit complex but it's becoming a real part of agent harness.

SkillOpt is one of the first papers to treat markdown skill files as trainable parameters and provides a proper optimization framework for them.

A few things I learned that you should consider too.

1. The validation gate is the only thing that matters in a self-editing loop.

Held-out set, strict improvement, ties rejected. End-to-end, their best skills land with 1 to 4 accepted edits total. If your "self-improving agent" is accepting most of what it proposes, you're shipping slop.

2. Bounded edits are better than full rewrites. 4 to 8 edits per step is the sweet spot.

Remove the budget and performance collapses. This is the textual analog of learning rate, and it transfers to any LLM-as-author loop. If you're using an agent to refactor your docs, your prompts, or your skills, cap the diff size.

3. Compactness wins. Median final skill: ~920 tokens.

Skills do not need to be long. They need to be high-signal. Most skill files I see are bloated because length feels like effort. It isn't.

4. The harness is becoming less important; the skill is becoming more important.

A Codex-trained skill ported into Claude Code hit +59.7 points on SpreadsheetBench. Procedural knowledge is more general than the runtime that
produced it.

5. Frozen model + trained context is the practical adaptation.

GPT-5.4-nano with a SkillOpt'd skill ≈ frontier behavior on procedural benchmarks. Cheaper, portable, inspectable, zero inference-time cost. This is
the answer to "how do we adapt a frontier model for our domain" for almost everyone who isn't training their own models.

6. Verification is the bottleneck.

Every gate in this paper depends on an auto-grader. That works for benchmarks. It fails for writing, design, and strategy, exactly the open-ended work we want to automate. Whoever builds the verifier for open-ended tasks owns the next stage.

There are also two leassons I learned while shipping v2.3.0 of my Context Engineering Agent Skills repo, measured across composer-2, claude-opus-4-7,
gpt-5.5, and gemini-3.1-pro via the @cursor_ai SDK:
- Description and body are two different surfaces. The router only sees the description. The agent sees the body once activated. They can quietly disagree, and only end-to-end task tests catch it.
- Aggregate accuracy is the wrong unit. When I rewrote three descriptions, the corpus average moved ~1pp. Individual skills moved 23–25pp. Per-skill effect size is where the action is.

Also, in Feb 2026 I shared a piece called Personal Brain OS arguing that the markdown file is a first-class substrate for agent state. SkillOpt is the optimizer-shaped version of that same argument: not "store memory in files" but "treat files as trainable parameters with proper optimization machinery around them." That's the move from static to measured.

The fast/slow split they describe already lives implicitly in the digital-brain-skill repo:
- voice-guide and tone-of-voice.md are slow-state (rarely touched)
- posts.jsonl and bookmarks.jsonl are fast-state

What SkillOpt adds that I didn't have is a protected section invariant, a structural guarantee that fast edits cannot overwrite slow lessons. Removing that mechanism cost them 22 points on SpreadsheetBench. Worth borrowing.

If you're building agents, SkillOpt: Executive Strategy for Self-Evolving Agent Skills is a good paper to read: https://t.co/ZS9SZXQ6Mv

[arindam @_arindam]

@arpit_bhayani Thanks. Will be there around July-Aug.

[arindam @_arindam]

@karpathy

[arindam @_arindam]

@arpit_bhayani Thank you for sharing. Fascinating redis internals

[arindam @_arindam]

@arpit_bhayani Happy Birthday 🎂

[arindam @_arindam]

Just realised I joined Twitter/X before Elon. Its been so long

[arindam @_arindam]

@arpit_bhayani Thanks for sharing. Redis is such an amazing design. Will watch.

[arindam @_arindam]

@arpit_bhayani Yes, facing this everyday. When you scale Agents, we fall back to basics of system design. The queues, DLQ, async comms between agents, redis, agent failures and restarts, checkpoints. Managing this at scale requires fundamentals. Its always core CS.

[arindam @_arindam]

Server sent (sse) streaming protocol is just magic

[arindam @_arindam]

@adamgordonbell What the ... Crazy

[arindam @_arindam]

@opentelemetry has such a brilliant design.
While tracing agents, it was an absolute treat to see the beauty of its abstractions, simplicity, and overall architecture. @Microsoft’s Agent Framework implements it in such an elegant way with built-in OTel trace emissions. Makes tracing and visualizing in @grafana via @Azure incredibly smooth.
I’m completely blown away! 🔥

[arindam @_arindam]

Ran into Boyar moore algorithm paper today, while searching for a faster string search algorithm. And whoa, it is implemented in grep!

[arindam @_arindam]

@Tigularius Hey!

[arindam @_arindam]

@RusUslada Hello!!

[arindam @_arindam]

@claudeai

[arindam @_arindam]

@beardwax2 Hey

[arindam @_arindam]

@birds_justice Really its like suddenly I can read mind of many people and see their thoughts.