Zil Patel

Making AI do the work.

[Demis Hassabis @demishassabis]

https://t.co/PTeDiv1b6L

[Zil Patel @zilppatel]

LangChain surveyed 1300+ engineers. Quality is the #1 barrier to agents in production. Not cost, not latency.

Most engineers can prompt an LLM into generating something that looks correct. The hard part is knowing when it's not. Subtle logic errors. Hallucinated edge cases. Code that passes review but fails in prod.

The skill gap isn't generation. It's evaluation. Knowing what to reject, what to rerun, and what to ship consistently at scale.

https://t.co/NFW3xkIN37

[Zil Patel @zilppatel]

@claudeai We've inverted software economics. Verification now costs more than generation.

A team shipping 30 PRs/week now pays $3,000/month to review AI generated code. Meanwhile CC max costs $200/month per developer.

[Zil Patel @zilppatel]

If you're reaching for an orchestration framework by default, try one model thinking harder first.

But know where that breaks: when the problem is hard enough that going deep on the wrong path means no recovery.

Code + results: https://t.co/hYrlC47pL2

What's your heuristic for when to add agents vs add thinking time?

[Zil Patel @zilppatel]

I ran GPT-5.4's extreme reasoning mode against a 3-agent pipeline on 9 real SWE-bench Lite bugs.

Same model. Same bugs. Different inference strategy.

The results challenge the multi-agent default most engineers are adopting right now.

[Zil Patel @zilppatel]

So the real tradeoff isn't speed vs accuracy.

It's: one model dynamically allocating reasoning vs multiple models providing redundancy when the problem is deceptive enough to be misleading.

Multi-agent = fault tolerance. Single model = peak performance.

[Zil Patel @zilppatel]

@simonw Strong agree on the unreviewed code dump. Beyond the individual side, do you think there's a team-level anti-pattern emerging here, like treating agents as 'free refactorers' without explicit ownership handoffs or review gates?

[Zil Patel @zilppatel]

The adversarial agent idea is solid. Bug finder pushes for everything, adversarial one fights back hard with scoring penalties, referee sorts it out. Clever way to get reliable bug checks without reading tons of code myself. Thanks for sharing @systematicls, stealing this for my CRs.

[Andrej Karpathy @karpathy]

Agency > Intelligence

I had this intuitively wrong for decades, I think due to a pervasive cultural veneration of intelligence, various entertainment/media, obsession with IQ etc. Agency is significantly more powerful and significantly more scarce. Are you hiring for agency? Are we educating for agency? Are you acting as if you had 10X agency?

Grok explanation is ~close:

“Agency, as a personality trait, refers to an individual's capacity to take initiative, make decisions, and exert control over their actions and environment. It’s about being proactive rather than reactive—someone with high agency doesn’t just let life happen to them; they shape it. Think of it as a blend of self-efficacy, determination, and a sense of ownership over one’s path.

People with strong agency tend to set goals and pursue them with confidence, even in the face of obstacles. They’re the type to say, “I’ll figure it out,” and then actually do it. On the flip side, someone low in agency might feel more like a passenger in their own life, waiting for external forces—like luck, other people, or circumstances—to dictate what happens next.

It’s not quite the same as assertiveness or ambition, though it can overlap. Agency is quieter, more internal—it’s the belief that you *can* act, paired with the will to follow through. Psychologists often tie it to concepts like locus of control: high-agency folks lean toward an internal locus, feeling they steer their fate, while low-agency folks might lean external, seeing life as something that happens *to* them.”

[Zil Patel @zilppatel]

There are no special tweaks on my end! Few things that might help:

• Make sure there is enough RAM (swaps to disk are pretty slow).
• Check if it's using CPU fallback instead of metal acceleration.
• Try the same model on different tools, I found LMStudio running the model faster.

[Zil Patel @zilppatel]

@ArixnCort @AlexReibman 16gb with the model using ~9.5gb

[Zil Patel @zilppatel]

@KevinSzabo14 One for service or giving back to causes you care about

[Zil Patel @zilppatel]

Building with AI is more approachable than ever for coders! What’s the best tool you’ve explored lately—lightweight LLMs, no-code platforms, or beyond?

Share your go-to tools!
#VibeCoding #LLM

[Zil Patel @zilppatel]

@Pauline_Cx I built MicroQuest - turning spare time into hyper-local adventures with tailored, bite-sized itineraries!
Try it at https://t.co/aZ4QppINJV

[Angel Perozo Prieto @angelppreneur]

Success is simple:

- Think long-term.
- Show up daily.
- Be obsessed.

The rest is noise.

[Zil Patel @zilppatel]

@karpathy Spot on! We're pivoting to LLM-first design and scrapeable data is key.

[Harsh Patel @harshpatel071]

Launching, join now:
https://t.co/xVXlCxTYea

#angular #nocode #v0 #bolt #loveable #startup #buildinpublic