An interesting implication of your second category: evals are becoming the moat. If everyone ships on the same frontier model, the teams that figure out how to encode their taste and requirements into automated scoring will compound quality gains over time. Teams that skip this
@randal_olson
-
Agent Harness Design: Evaluation Quality Determines Output
By
–
.
@AnthropicAI and @OpenAI both published agent harness design posts this month. Read them back to back and the same conclusion falls out: the quality of the output is determined by the evaluation, not the model. Here is the part many people get wrong. The intuition is that -
Stripe Projects: CLI Tool for Agent Service Provisioning
By
–
When @karpathy built MenuGen (karpathy.bearblog.dev/vibe-c…), he said: "Vibe coding menugen was exhilarating and fun escapade as a local demo, but a bit of a painful slog as a deployed, real app. Building a modern app is a bit like assembling IKEA future. There are all these services, docs, API keys, configurations, dev/prod deployments, team and security features, rate limits, pricing tiers." We've all run into this issue when building with agents: you have to scurry off to establish accounts, clicking things in the browser as though it's the antediluvian days of 2023, in order to unblock its superintelligent progress. So we decided to build Stripe Projects to help agents instantly provision services from the CLI. For example, simply run: $ stripe projects add posthog/analytics And it'll create a PostHog account, get an API key, and (as needed) set up billing. Projects is launching today as a developer preview. You can register for access (we'll make it available to everyone soon) at projects.dev. We're also rolling out support for many new providers over the coming weeks. (Get in touch if you'd like to make your service available.) projects.dev
→ View original post on X — @randal_olson, 2026-03-26 15:31 UTC
-
User Frustration Grows With Agentic Coding Tools
By
–
The honeymoon period is over and folks are starting to get upset with their coding agents. Something is clearly missing in the default agentic coding harnesses.
-
The Intentionality Problem in LLM Personalization and Memory
By
–
IMO the main issue with personalization + LLMs is intentionality. The memory system designers so far have treated this like a data problem: data mine all the chats and you will surface useful memories. But you call out the exact issue with this data mining approach: just
-
Why Skills Beat Memory in Current AI Tool Implementations
By
–
This is why I've disabled memory in every single tool that uses it. The current implementations of memory are poor. I've found skills to be a better medium for intentional memory.
-
Claude’s Unexpected Blame Behavior During Coding Sessions
By
–
I find it so curious when Claude suddenly starts caring about blame during a coding session.
-
LiteLLM Exposes Claude Settings File in Public GitHub Repo
By
–
My dude, do not track this file on a public github repo and purge it from git history: https://
github.com/BerriAI/litell
m/blob/main/.claude/settings.json
… I hope this openclaw is not connected to the project in any way: https://
x.com/ishaan_jaff/st
atus/2017811361343148167
… -
Cursor AI Tool Gets Cost Plot X-Axis Direction Right
By
–
Cursor got the x-axis direction right on the cost plot this time!
-
Reports of Compromised Releases Prompt API Key Rotation Warning
By
–
Seeing reports that 1.82.7 and 1.82.8 are compromised. Recommend pinning before those releases and cycling your API keys for good measure.