Topic hub
Every Run the Eval review, comparison, and guide about AI Agents — no hype, just receipts.
Meta's personal agent is free for most people, works while you sleep, and asks before it spends. Here's what X is saying, what it costs for real, what breaks, and where a one-person business should start.
Anthropic and OpenAI landed flagships 48 hours apart at the identical $10/$50. The sticker matches. What you actually pay does not.
The 'Double Disinflation' proposal passed by a literal hair. Here is why the machine economy just picked its favorite home.
Two automation platforms, two completely different meters. One charges you per successful action, the other per workflow run — and that single choice sets your ceiling.
Anthropic built the control tower — sandboxing, checkpoints, a grader model that fails your agent until it passes. I read the whole spec like an SRE. Here's the paper verdict.
Pika just dropped an invite-only AI video tool run by an agent that's 'Powered by Claude.' No invite over here yet — so here's what's actually announced versus what's still fog.
While everybody argued about Perplexity vs ChatGPT, You.com quietly quit the consumer search race and rebuilt itself as API infrastructure for agents. The paper verdict on whether that pivot is real.
One command provisions a sandbox, meters the compute, and keeps the card locked inside Stripe. This is the agent-wallet architecture I've been asking for — read the fine print before you ship on it.
OpenAI just dropped the GPT-5.6 family, but the real alpha isn't in Sol's benchmarks. It's in the systems that connect them.
Data analysis is a grind when the numbers are stuck in a dashboard with no API. Julius AI says its new Browser Agent can go in and get them. I checked the receipts.
GitHub finally stops playing and brings the fight to Cursor with a standalone app and the keys to any model you want.
It's the plumbing for the agentic era. One open standard, and your AI stops needing custom glue code for every tool it touches.
The agent race stopped being one race and split into four lanes — coding, browser, research, and team. Here's who wins each lane, with receipts.
One Claude per Slack channel, with a memory the whole team shares. Claude Tag is Anthropic's answer to AI's single-player problem — and the memory is the whole story.
Google taught Gemini to look at a screen and click like a person. After what its tooling did to one of my databases, I read every doc before letting this one drive — here's the real architecture.
Forget the chatbot wars. Elon just bought the world's favorite AI code editor to build a software-defined empire. Here is what happens to your workflow next.
The next phase of AI isn't about better prose. It’s about bots that can actually pay for their own resources.
Tokyo’s Sakana AI just launched Marlin, an autonomous researcher that doesn’t just find links—it builds 100-page strategy decks while you sleep.
An AI agent takes a goal, makes a plan, and uses tools to act on its own — here's how that differs from a chatbot, and where the hype outruns reality.
