The archive
Systems that plan, use tools and act on your behalf.
OpenAI's customer story claims invideo's agentic video editor tripled its color-grading success rate and built about 50 custom effects in a day. The baselines behind those numbers were not disclosed.
5 min read
OpenAI says GPT-6 ships with a reworked prompt caching system, a 30-minute reuse window, a caching dashboard and miss diagnostics — with customers reporting hit rates above 90%.
5 min read
OpenAI's customer story credits GPT-6 Astra with cutting Parallel's research agent runtime and cost roughly in half. The numbers are relative, the methodology is thin, and the baseline is unnamed.
5 min read
V7 turns enterprise documents into a queryable graph of entities and cited evidence, then serves it to agents over MCP. The architecture is sound; the vendor's own numbers need a careful read.
6 min read
AWS and Benchling describe a multi-tenant sandbox for agent-written code that refuses DNS resolution by default, scopes storage access per job, and wires exfiltration tests into CI.
6 min read
A plain-language definition of AI agents: the loop that powers them, the parts they are built from, and how they differ from chatbots that only answer questions.
8 min read
A practical framework for placing people inside AI workflows: where to require approval, how to escalate, and why reversibility should drive oversight decisions.
8 min read
A step-by-step approach to evaluating AI agents: building task suites, defining success criteria, reviewing trajectories, and catching regressions before users do.
8 min read
How multi-agent AI systems split work between orchestrators and workers, how handoffs pass context, what they cost, and when a single agent is the better choice.
7 min read
How language models call tools: writing schemas, running the request-execute-return loop, handling errors, and scoping permissions so models act safely.
8 min read