Neel

[The Shift Journal @TheShiftJournal]

Did you know?

[Neel @neeilk2028]

@_christinacodes I would rather know about the person , and why he has the pov , before than sit through the entire session only to find out it wasn’t worth it. A “brief” intro with context is not at bad thing, IMHO. Helps me place the person and many times make him relatable.

[Huberman Lab Clips @HLPClips]

"A professional shows up every day, stays on the job all day, does not take success or failure personally."

"An amateur worries about how they feel. A professional doesn't care how they feel. They do it."

"All you have to do is say to yourself, I'm going to attack this thing as if I were Kobe Bryant."

"Turning pro does have a cost. A lot of times you have to leave people behind."

"Groups of friends will have an unspoken compact among them that we're all going to stay mediocre. That's the deal."

@spressfield on @hubermanlab

[Matt Pocock @mattpocockuk]

Everyone always confuses my skills with spec-driven-development.

It really annoys me. The specs my skills create are intended to be deleted immediately - not kept around, or treated as source code.

Birgitta Boeckeler calls this 'spec-first' development, but still groups this under the SDD umbrella (link below).

I disagree - I think it needs its own term. The specs aren't that important. They're just a projection of the decisions made during grilling.

GDD? Grill-driven-development. IDK

[Neel @neeilk2028]

@PoorserviceLDN @CoachJulianNiko Would you rather not Spend 45mins and check if it works for “you”….you need a study by someone else? What if the study is wrong and they come to correct it after another study disproves it?

[Neel @neeilk2028]

Hmmm

[Andrej Karpathy @karpathy]

One pattern I find useful for working with LLMs is a nice long ramble session. Sometimes the LLM needs more bits to understand what you're trying to achieve, but you're too lazy to type them. In these cases I like to lean back, switch to /voice and just ramble for like 10 minutes, total mess, anything goes, full stream of consciousness. Sometimes I declare it up top, something like "switching to speech recognition sorry for any typos...". Sometimes I turn it into a small interview of a few turns. But I find that the LLMs are somehow very good at reconstructing long incoherent rambles and often their echo of your own tangle of thoughts comes out quite a bit cleaner than what you started with. The result is that you improve the mind meld and have to correct things less from that point on.

[Gergely Orosz @GergelyOrosz]

Knowing how LLM contexts work and how to work around context limitations – aka “context engineering” – is becoming so important. No better person to explain than @dexhorthy

Timestamps:

00:00 Intro
01:33 Dex’s path into tech
03:34 Early work in platform engineering
05:28 Replicated
11:24 Metalytics
12:36 12-factor agents
18:27 Context engineering
23:38 Harness engineering
26:11 Context overload
30:45 Loop engineering
44:34 Software factories before and after AI
50:33 Automation limits
55:18 Three options for automating
59:00 RPI framework
1:04:16 Intentional compaction
1:11:48 Token harder vs. token smarter
1:16:44 AI slop
1:19:15 HumanLayer
1:29:09 Book recommendation

Brought to you by:

• @AntithesisHQ — with Antithesis, you can use AI agents to work on critical systems without worrying about correctness. Teams like Jane Street, https://t.co/MTEw4IA0qA, and the etcd community use Antithesis to ship better code, faster. https://t.co/AKYm4cbVCU

• @buildkite  — the CI orchestration platform built for reliable scale. Used by OpenAI, Anthropic, Cursor, Meta, Uber, Ramp, Nvidia, Airbnb and many more. https://t.co/NOnTAHsCDD

• @sentry — application monitoring software built by developers, for developers. Check out their AI agent, Seer AI, and Sentry MCP. https://t.co/Ig2WcWttl3

Three interesting learnings from this episode:

1. Lesson learned: Shipping unread code spells disaster within months.

Dex experimented with having the model write the code and humans not reviewing anything in July 2025. Four months later, they shut things down and threw the whole system out. Production broke, and no matter how much the team prompted Opus 4.1, the model could not find the root cause. Once fixed, it took three weeks (!!) to re-onboard to a codebase no human had ever read

2. Context engineering 101: figure out where the “dumb zone” begins.

As a rule of thumb, the less of the context window that is used, the better the outcomes are. This is because the attention mechanism is quadratic: the more that goes into the context window, the more compute is required to process it all.

3. “You’re completely right!” or “you’re right to push back on that” are phrases that mean it’s time to start a new session.

These responses mean the LLM session is trajectory-poisoned, and you’re wasting time and tokens to continue. This is because models are autoregressive.

[Neel @neeilk2028]

@esskayesss_ Why not?

[The All-In Podcast @theallinpod]

Brad Gerstner: Companies Will Pay 5x More for the Best AI, No Evidence of Pricing Pressure from Open Source

@altcap:

“Jason, you talked about summarizing a document, it may take 20,000 cheap tokens to do. Of course, shoot that to a lagging model or an open source model.

But if you're talking about replacing a software engineer for two hours, that may take two million expensive tokens, and the consequence of using something that's 95% as good is really high.

Because you have a long-running task, and if the task breaks early, or it breaks in the middle, or it breaks at the end, there's a huge cost to that.”

@Jason:

“You still burn the tokens, right? And back to this analogy I was using, you're pulling the slot machine, and you lose.”

Brad:

“And (you lose) the time and the compute.

So if an AI agent is replacing a $200 an hour consultant, right? Take that as an example.

So three consulting firms, they're competing. They need the smartest consultant. They're charging $200 an hour.

The difference between spending $3 on a cheap model or $15 on an expensive model to replace a $200/hour consultant, it's just irrelevant.

That inference cost difference is irrelevant if you're getting something that's bulletproof for $15, and so I think that's what we're seeing play out.

The best evidence for all of this is just revenue growth. I'm talking about, what is Anthropic's revenue growth compared to OpenAI, compared to the open source models? Millions of independent actors are choosing every single day.

The open source companies are growing, right? But they're growing selling something that is really, really cheap.

And there's room in every single market for premium products, for mid-tier products, and for commodity products, and I think we see a lot of this token growth, people are speculating that the intelligence gap between that commodity stuff and the frontier stuff is going to collapse to the point that people won't pay for the frontier stuff.

There is no evidence of that on the field today.”

[Martin Fowler @martinfowler]

NEW POST

Birgitta Böckeler recently spent some time trying out running local LLMs for some programming tasks. In this memo she outlines the factors that influence how viable they are for the job.

https://t.co/ALp5q6p3g5

[Mitchell Hashimoto @mitchellh]

Mind boggling to me that I can make a thing faster and there's always people that ask "but why?" What kind of mentality is that? The pursuit of excellence does not need justification. Also, I find in so many cases, we can't know the impact of an improvement until we do it.

For example, one I've talked about before: Ghostty's high IO throughput has enabled terminal program (emulator and TUI) fuzzing at a speed thats incomparably fast to prior solutions. This has resulted in upstream patches to resolve issues in popular projects like btop, tmux, and more.

Speed enabled that anecdotally example that lifted the tides of adjacent communities that don't rely on Ghostty technology at all. I didn't predict this.

Make things better because they can be better and let the results naturally play out.

[stevibe @stevibe]

You know that "But, wait..." moment in every LLM thinking trace?

I made it visible.

I asked 8 models the same tricky probability question and rendered their reasoning as trees. Every time a model rejects its own idea and pivots, every "But...", every "Wait, actually...", a new branch grows.

Same question. Completely different minds.

[Neel @neeilk2028]

@ttorres I’m sorry you had to go through this.

[Neel @neeilk2028]

@hopperlife @nikitita444 @SawyerMerritt @Scobleizer Or he fainted ??

[Neel @neeilk2028]

@thatsmypjamz @TrungTPhan Folks try reduce Suffering by sharing .. or commensuration

[Reads with Ravi @readswithravi]

“You escape competition through authenticity.”

— Naval Ravikant

[Garry Tan @garrytan]

For people who don’t have a clear sense of the future they want, AI is just another mechanism of control

But in the hands of someone with agency, AI is the breaker of chains, something that lets you do things no humans can do alone.

AI can be a liberator if you choose agency.