Aidin Khan

20 y/o reverse-engineering why AI agents fail. Autopsies, evals, fixes · so you debug faster than I did

[Aidin Khan @AidinRoR]

@BraedendotTECH reminds me "if everyone is special, no one is"

[Aidin Khan @AidinRoR]

Agentic evolution is driving toward less human control, look:

1. Static (question → answer) → branch flow(LangChain), prompt engineering (ReAct, Chain-of-Thought), rag
2. Graph workflows -> state, cycles, process automation (LangGraph, PydanticAI, n8n, Dify)
3. Agentic harness -> AI personal assistants, coding agents (Claude Code, Codex, OpenClaw, Hermes)
4. Loop engineering -> triggering the harness above

What's next?

[Aidin Khan @AidinRoR]

@yadavji_codes Need to build a brand around the fact that you made $1M using only your MacBook and Claude)

[Aidin Khan @AidinRoR]

@mercurithinks @kimmonismus or a genius who understands marketing

[Aidin Khan @AidinRoR]

@pankajkumar_dev They are doing it just as Anthropic is pulling Fable from the subscription)

[Aidin Khan @AidinRoR]

@_StanGirard @AnthropicAI Fable feels insanely good at frontend development and at natively understanding what I want. But I do think if codex keeps up on frontend, I think using only Codex is enough for most of people needs

[Aidin Khan @AidinRoR]

@IamVishwajeet99 @sahill_og If you are good at calculus, probability, statistics, it will benefit you a lot

[Aidin Khan @AidinRoR]

Software engineers are so far away from dying.

YES, Fable or any new frontier AI makes it easier for more people to produce code and understand it. Therefore, YES, it is easier to build and ship products, at least mvps.

But in a real job, there are too many skills needed: problem solving, abstract thinking, trade-offs, debugging, architecture, maintenance, fucking judgment, etc.

I think when a job is truly dying, people usually stop thinking about it.
But SW is discussed more than ever.

[Claude @claudeai]

Fable 5 is back.

[Yingjun Wu 🤘 @YingjunWu]

Honestly, I've been losing interest in building AI agents over the past two or three months.

Like many engineers, I got incredibly excited about coding agents since December 2025. They fundamentally changed the way I work. I could not only dream bigger, but also build those ideas myself with tools like Claude Code.

So why did I lose interest?

One reason is that I struggled to see much real novelty, or so-called "moat". Everyone seems to be building roughly the same products, using the same models to create slightly different workflows for marketing, sales, content generation, and other use cases. I felt bored with all kinds of "new products."

But the bigger reason is that it was frustrating to see so much work go to waste. Many projects stopped at the demo stage and never made it into production!! 🟥

How many agentic workflows are actually running inside real businesses? How many customer support agents are handling real traffic from frustrated customers Production systems have almost 0 tolerance for failures, incorrect tool calls, or inappropriate responses. Yet AI agents, by their nature, cannot provide absolute guarantees.

That is why I've returned to the "boring" world and resumed my development of AI/data infra as well as foundation models.

The joint work (to be announced soon!) between @RisingWaveLabs and @ArklexAI aims to power the next generation of agentic AI for production-grade enterprise workloads.

Stay tuned!

[Aidin Khan @AidinRoR]

@ishanxtwt bc front software engineer can vibe code better

[Aidin Khan @AidinRoR]

@ChShersh

[Aidin Khan @AidinRoR]

@ManOfFocus_ they are smart, smart enough to move fast, take risks, work a lot