How Anthropic's Dynamic Workflows Coordinate Claude Agents

At Monday's workshops, Anthropic drew on lessons from Unix and Linux to show how its new dynamic workflows, or "flows," let Claude Managed Agents take on bigger multiagent projects without overwhelming the developers overseeing them.

AI Engineering Has Entered Its Audit Era

An analysis of the roughly 140 sessions at AI Engineer New York. Two in five have "agent" in the name, but the more telling shift is what people now want to do about agents: verify, govern, and audit them.

What 1,368 Real-World AI Incidents Teach Us About Evaluation

Evals have gotten sophisticated at testing models, but a study of real-world AI failures found that many of the ones that mattered most weren't model failures alone. They emerged from incomplete data, poorly designed workflows, weak monitoring, or no one able to intervene.

Kepler's Vinoo Ganesh on Building AI You Can Trust

Kepler's CEO, a Palantir and Citadel veteran, talks about building AI for industries where accuracy matters, why Kepler's model never writes a number, and why a bigger model isn't the answer.

What The Rebirth of Classifiers Tells Us About ChatGPT's Success

Jev, a promptable classifier, took X and the enterprise by storm in September, and OpenAI, Cloudflare, and Liquid quickly followed with their own. What its rapid adoption reveals about why ChatGPT succeeded in the first place.

AI Needs a Test That Fights Back

One right answer is not the same as a good sequence of decisions. Unlike a fixed answer key, an opponent responds, and eSlams is building a public stadium to test how AI handles that.

Code is abundant. Confidence is a formula.

Agents can now produce an initial solution in minutes, and the volume of code will only rise. The next challenge for engineering organizations isn't generating more code; it's creating confidence in the code that reaches production.

Brakes Let You Go Fast: Balancing Acceleration and Safety for AI Agents

At a regulated firm like Two Sigma, too little safety is an existential risk and too little innovation is a slow death. How the right controls let agents operate in production without handcuffing what they can accomplish.

Frontier AI Models Outperform Human Experts on Earnings Prediction for the First Time

Samaya built a finance-specific prediction harness to test whether frontier AI can beat experts in one of the hardest areas of knowledge work. On earnings prediction, the models came out ahead.

Workshop Day Opens AI Engineer NYC

Monday's 18 hands-on workshops put attendees at their keyboards from the first session, and the lineup doubled as a snapshot of what AI engineers are working on right now. The Expo floor and a welcome reception closed out the day.