#reasoning
Every post tagged "reasoning".
13 posts · showing 1–10
-
The Travel Agent Reviewed Its Itinerary Three Times and Approved the Wrong Date Every Time. Reflection Needs Something to Check Against.
A self-review loop that reads only its own draft can polish the wrong answer forever. The fix is to give the critic something the generator never had: the original request, a fresh context, or a tool that can say no.
-
Your Agent Called the Same Tool Three Times in a Row. Here's What the Loop Is Telling You.
A repeating agent isn't being stubborn — it's being consistent. Identical tool calls mean the context barely changed between turns, and there are only four reasons that happens.
-
Four Tool Calls to One Answer: A Research Run, Annotated
A research agent answers one question in four steps. Reading what changed in the transcript between each step explains more about how agents work than any diagram of the loop.
-
Novel Problems vs Recombined Ones: The Line That Matters
On genuinely novel problems you don't get an error — you get the nearest known pattern applied confidently to an adjacent problem. One classification decides where your attention belongs.
-
Ambiguity Resolution Stays a Human Job for a While
An agent doesn't fail on ambiguity — it absorbs it, picks a reading, and produces a confident answer to a question nobody asked. Better models guess better; they don't guess less.
-
Why Agents Struggle Where There's No Feedback Signal
Coding agents work well because compilers exist, not because code is easy. Effectiveness tracks whether a domain has a fast, independent, specific verdict — and that's frequently something you can build.
-
Taste Is a Bottleneck: Where Judgment Has No Substitute Yet
Cheap generation doesn't reduce the value of judgment — it makes discrimination the scarce step. And it introduces a hazard: well-argued cases for mediocre options, which is what agents produce uniformly.
-
Agents Are Better at Writing Code Than at Deciding What to Build
Both feel like thinking about the problem, and only one has complete inputs and a checkable answer. The reframe that makes a model genuinely useful near a decision it shouldn't be making.
-
Getting an Agent to Say How Sure It Is
Ask a model how confident it is and you get a plausible number that doesn't move with difficulty. Five signals that actually correlate with being right — and the calibration check that tells you if yours is noise.
-
The Verifier in the Loop: Designing an Oracle Your Agent Can Iterate Against
Coding agents work as well as they do because compilers exist. In other domains the oracle is something you build — and most tasks have one hiding in them if you spend ten minutes looking.