The AI Record
if you're just getting started with AI, try to keep it around 100,000 tokens. For larger million context window, we … this up to … 200,000 tokens, but I've regularly tried to keep it under 60 for the hardest problems. I've regularly gone over 300K for things where I'm just kind of riffing with the agent, and I'm just … too lazy to compact it and move on and do a new one.

Dex HorthyCEO, HumanLayer · Executive, HumanLayer

Context. Horthy is answering the moderator's question on whether context rot and context engineering are solved now that context windows are larger.

Talk: The Great Loops Debate, AI Engineer (YouTube), July 17, 2026, at 28:36.

  • 100,000 tokens context guideline new users, as the speaker states it
  • 200,000 tokens context guideline large windows, as the speaker states it

Checked against the source [ Grade B ] Said in the last year Sells AI tools

Filed under

Who is speaking

Dex Horthy, CEO, HumanLayer. Runs the company whose coding-agent practices he describes; HumanLayer sells AI coding tooling.

HumanLayer sells AI or agent tools or services, so it has a reason to publish flattering numbers about AI. The passage stays in the archive; read it with that in mind.

How this was checked

Checked against the source by a first reader, then confirmed independently by a second reader who fetched the source without seeing the first reader's notes (Independent review agent (Claude Sonnet), blind second pass from passage text and citation only, no repository access, 2026-11-04).

Whisper transcript at about 28:36–28:58. The passage's ellipses cover 'provide', 'like' and 'like', and 'just like too lazy' becomes 'just … too lazy'. No captions.md exists for this video. The speaker is the debater introduced as Dax Horti (CEO of HumanLayer); whisper misspells the name, but the cadence and content fit Dex Horthy.

How to cite this passage

Dex Horthy (HumanLayer), “The Great Loops Debate”, AI Engineer (YouTube), 2026-07-17, at 28:36. https://www.youtube.com/watch?v=c35YoMdnI78 Archived at https://the-ai-record.com/passages/humanlayer-context-window-guideline-100000-tokens-under-60-hardest-problems-2026

More from the same source

And so my advice is … instead of trying to automate everything end-to-end, build these small incremental loops throughout your system, and you will wake up one day and you will be moving two to three times faster while still being able to read the code, while still owning the architecture.

Dex HorthyCEO, HumanLayer · Executive, HumanLayer

Context. Horthy is answering the moderator's question on how to reach fully autonomous software factories if two to three times is what coding agents give most engineers.

Talk: The Great Loops Debate, AI Engineer (YouTube), July 17, 2026, at 48:33.

  • 2 x speedup low, as the speaker states it. A person approves
  • 3 x speedup high, as the speaker states it. A person approves

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

I was playing with Notion and we used Notion a lot at Keycard, but I really just wanted all my Notion things available to me as Markdown files. And because … just made it easier for the agent to work with it instead of going through MCP

Ian LivingstoneCEO and co-founder, Keycard · Executive, Keycard

Context. Livingstone is answering the moderator's question about how to control access to a memory store shared by many agents.

Talk: The Great Loops Debate, AI Engineer (YouTube), July 17, 2026, at 40:42.

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

I need to say, it has been 10 months now, I do, I minimally use any open source software, minimally. I just generate it to my requirements. And then when a supply chain attack happens, I'm like … affect me.

Geoffrey HuntleyCreator of the Ralph loop; Latent Patterns · Practitioner, Latent Patterns

Context. Huntley is giving his closing statement after the moderator and another panelist raised software supply chain security.

Talk: The Great Loops Debate, AI Engineer (YouTube), July 17, 2026, at 58:39.

  • 10 months months minimal open source use, to July 2026, as the speaker states it. Level not stated

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

And I don't see myself going back to writing code by hand. It's been two and a half years since I manually wrote code by hand.

Geoffrey HuntleyCreator of the Ralph loop; Latent Patterns · Practitioner, Latent Patterns

Context. Huntley is speaking in his opening four-minute statement for the side arguing that loops work today, in an Oxford-style debate.

Talk: The Great Loops Debate, AI Engineer (YouTube), July 17, 2026, at 5:31.

  • 2.5 years years since hand written code, to July 2026, as the speaker states it. Level not stated

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

Something I've done for a long time, and it's very silly, is I run a model … without any skills or any markdown. … actually, I get rid of all my skills and all my markdown and everything when the new model is released.

Geoffrey HuntleyCreator of the Ralph loop; Latent Patterns · Practitioner, Latent Patterns

Context. Huntley is answering the moderator's question on how he keeps a model from cheating when verifying its own work.

Talk: The Great Loops Debate, AI Engineer (YouTube), July 17, 2026, at 32:35.

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

So the concrete example here is we do security scanning after on our PRs in local and after our PRs even land because they will always find some things that are real that we have overlooked. And they sort of beat humans on the … code review. And it's expensive. It … us, I think like five bucks a PR or something like that to run all the checks that we want. But that's where we made an explicit decision that it's worth it.

Greg PstruchaDeveloper, Sentry · Practitioner, Sentry

Context. Pstrucha is answering the moderator's question: 'when does a loop pay for itself, and how often is this actually the case?'

Talk: The Great Loops Debate, AI Engineer (YouTube), July 17, 2026, at 36:59.

  • $5 cost per pr checks, per pull request, as the speaker states it. Level not stated

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

Something wrong? Report a correction. JSON: /api/v1/passages/humanlayer-context-window-guideline-100000-tokens-under-60-hardest-problems-2026.json