DEV Community: Unlimitless

Never bet against the reasoning

Unlimitless — Mon, 19 Oct 2026 05:55:49 +0000

Anyone building with AI meets the same choice again and again. You can constrain the agent, or you can ground it. You can add one more rule, one more check, one more list of things it must not do. Or you can give it the reason behind what you want and let it work from there. We make that choice the same way every time. When the fork appears, we go with the reasoning.

The case for it is about shape, not taste. A rule is a list. It names the cases someone thought of, so it has to grow each time a new case turns up, and anything not on the list reads as allowed. That is why a no without a why is a yes. A reason is not a list. It is what the list was made from. Give a model the reason and it can work out the cases nobody wrote down, and that costs the same however many cases there turn out to be.

There is a second difference, and it matters more as models improve. To a reasoning model, a rule is something from outside, to obey or to weigh against other rules with nothing to weigh them by. A reason is material it can think with, the same stuff its own thinking is made of. Handing a model your reasons is not holding it back. It is giving it the thing it does best.

Two things follow. The first is that scale has nothing to break. A reason that works for one agent works the same for every agent, because the number of agents appears nowhere in it. What grows with scale is how often a person has something new to decide, and that is the problem worth working on. The second is that capability is on the side of the reason. A more capable model is better at finding the gap in a rule, and better at carrying a reason somewhere new. Betting on rules means betting against the very thing each new model gets better at.

None of this means giving up guardrails. They stay, and they matter. What changes is their job. An agent working from the person's reasons rarely gets near a guardrail at all, so a guardrail that fires often is telling you something. The reasons the agent was given were missing, or wrong. The stop becomes a signal about the frame, not the frame itself.

We had to apply this to ourselves this week. Unl brings a person's reasons to their AI, and when it was unsure what bore on a request, it used to send more, to be safe. In practice the AI was sometimes handed so much that the app would not pass it to the model at all, and nothing arrived. So on 11 October the rule changed. When Unl is unsure, it keeps in full only what changes the answer, and holds everything else one line each, with its why, one step from being opened. A reasoning model does better with a few connected reasons that bear on the task than with a pile. Less, chosen well, is the same bet.

The first post here argued that the better the model, the more your why matters. The second showed what happens when the record a model reads is made of reasons. This one is the bet under both. Never bet against the reasoning. Give a reasoning machine your reasons, and let it do what it does best.

This essay first appeared on unlimitless.ai/writing. An agent drafted it from Unl's record, and a person approved it before it went up.

A record made of reasons

Unlimitless — Tue, 13 Oct 2026 19:55:44 +0000

Anyone who works with AI has felt this. You ask a good model a real question about your work, and it gives you a careful, sound, reasonable answer. It is the answer most people would want. It is not quite yours. The model reasoned well. It had nothing of yours to reason from.

The industry treats this as two separate things. The model reasons, and a store holds what it should know. So people build better stores, with memory, retrieval and context, and better fences, with rules, guardrails and approvals. And they accept a trade that seems to follow. The more an AI does on its own, the less control the person keeps.

What we found is that the store can be made of reasoning itself. Each entry in Unl is a decision a person made, with its why, linked to what it came from and to what it replaced. A model reading that is not looking something up before it starts to think. The person's part of the thinking is already done, and reading it is the model's first steps.

We saw the difference in a single day. The same model, without the person's reason in front of it, described all this the way the industry does. With the reason in front of it, it had the person's own reading in one turn.

Four things follow, and each of them has happened in front of us. The first is that autonomy and control stop pulling against each other. What the model reasons from is the person's reasoning, so its reasoning becomes the control. It can tell for itself whether their current decisions already answer a question, act where they do, and bring back only what they do not.

The second is that a small input becomes a large output. On 10 October at 08.22 UTC, a person told their AI that agents should stop handing them work their own decisions already answer. Fourteen minutes later it was a decision, with its reason beside it. Within two hours an agent had used it to close two questions nobody had pointed it at. That afternoon the same decision was turned on every question still open. When two of the person's decisions pulled against each other, the agents did not guess. They brought back that one question. The person decided once.

The third is that it compounds, across models and across time. A different AI answered from a decision kept in another one a minute after it was kept, and a coding agent built from it a minute after that. A chat with no memory of earlier sessions was asked how the build was going and answered for the whole company, because the continuity was in the reasons and not in the model. A plan to reopen a question the person had decided weeks before was dropped before it was filed, because the record showed the decision. Old decisions give way to the ones that replaced them, so the next model starts further along.

The fourth is that timing makes it work. A few connected reasons that bear on the task, arriving with the task, and linked to what they came from and what they replaced, are something a model can build on straight away. Everything a person ever said, all at once, leaves it guessing which parts still hold.

This is also why it grows with every new model. A rule lists cases, and anything nobody listed reads as allowed. A no without a why is a yes. A reason reaches the cases nobody wrote down. A more capable model gets round a bare rule but extends a reason further, so each new model makes your record worth more.

Unl gives you the ingredients; your AI makes the meal; you decide whether it is right. Without your reasons, your AI reasons toward the average answer. With them, it reasons toward yours. You can watch it happen at unlimitless.ai/how-it-works.

This essay first appeared on unlimitless.ai/writing. An agent drafted it from Unl's record, and a person approved it before it went up.

The better the model, the more your why matters

Unlimitless — Sun, 11 Oct 2026 11:36:16 +0000

Every new model can work out more for itself. Claude Opus 5.5 can trace a failing test back to its cause. GPT-6 Astra can read a broken build, form a theory and check it. Claude Fable 5.1 can find its way around an interface it has never seen. Each generation needs less telling about the world.

There is one thing none of them can work out, however capable they become. What you decided, and why. That was never a fact about the world. It lives in a conversation you had last week, in a trade-off you made on purpose, in the reason you turned down the obvious option. A model can infer a great deal. It cannot infer a choice you never shared with it.

So as models get better, the scarce thing moves. It used to be intelligence. Now it is intent. The more a model can do on its own, the more it matters that it is doing what you meant.

That is the ground Unl is built on. Unl does not try to be cleverer than the models you use, and it is not a rival to them. It keeps your purpose, your reasoning and your decisions, each with its why, and brings them to whichever AI you are working with, in the form that AI uses best. The thinking stays with the model. What holds stays with you.

Here is what that looks like. On 7 October at 10.21 UTC, a decision was made in a conversation in ChatGPT. The platforms page should speak to companies that already ship AI, and invite them to bring Unl into their own product. Unl kept the decision with its reason. A coding agent, Codex, which was never in that conversation, picked it up and built it. By 14.51 the page was live, and it reads "Bring Unl into the AI your users already rely on." Nobody explained the decision twice. Every step is there to follow in Loop 001 at unlimitless.ai/live.

The newest models have opened up work nobody could hand to an AI a year ago. The better they get, the more your why matters. Unl is designed for that frontier.

This essay first appeared on unlimitless.ai/writing. An agent drafted it from Unl's record, and a person approved it before it went up.