ISSUE 512
50 experiments, 0 humans, 1 wild week

"At some point, you need to make a serious choice: modernize your analytics so you can compete faster, or risk becoming irrelevant."

— Tower CEO Serhii Sokolenko. Read more here.

OutSystems CEO on how enterprises can successfully adopt vibe coding

Everybody is building AI agents. But the enterprises actually shipping them to production are learning that an agent on its own is rarely the answer. In this episode of The New Stack Agents, OutSystems CEO Woodson Martin explains why AI agents alone fail in production and how blending agents, APIs, workflows, and human oversight drives real ROI.

Week in review: 50 experiments, 0 humans, 1 wild week

The story that dominated our traffic charts this week at The New Stack was born in an unlikely place: a 630-line Python script. This report by Janakiram MSV is about Andrej Karpathy’s AutoResearch, which ran 50 AI experiments overnight on a single GPU without any human input.

The second-most read story of the week was Paul Sawers’ report on Anthropic quietly removing the long-context pricing surcharge for Claude Opus 4.6 and Sonnet 4.6. That change means 1-million-token context windows are now available at standard per-token rates. Also from Paul this week is a look at Tower, a startup founded by ex-Snowflake engineers who want to tackle a blind spot in data engineering.

Our story on how MCP’s four biggest production-use growing pains will soon be solved was also widely read. See what may be fixed in 2026.

Finally, Jessica Wachtel's beginner's guide to vibe coding held its own, boasting the longest average read time of any article in our top five. 

See you next week.

— Nick Lucchesi, editor-in-chief, TNS

OpenAI’s GPT-5.4 mini and nano are built for the subagent era
The new models approach GPT-5.4 on coding benchmarks at a fraction of the cost, and they're designed to be delegated to, not used alone.

GitHub Copilot’s effect on collaboration has stunned researchers
A Harvard study of 187,000 developers finds GitHub Copilot reshapes how programmers work, boosting coding time 12.4% while cutting project management by 24.9%.

Managed OpenClaw bids to kill the hidden token tax on AI agents

Featherless launches Managed OpenClaw, a serverless AI agent platform with flat-fee inference to cut token costs and DevOps complexity for developers.

Nvidia brings together AI labs to build the next generation of open base models

The Nemotron coalition brings together several AI startups to collaborate on a new base model that will serve as the basis for Nemotron 4.

Why AI-driven operations are pushing governance beyond a compliance issue and into an operational priority

The Nemotron coalition brings together several AI startups to collaborate on a new base model that will serve as the basis for Nemotron 4.

Building production-ready agentic AI
March 24

Generative models alone aren’t enough for production-grade AI. Without a unifying control plane, scaling autonomous agents often leads to hallucinations, cascading errors, and unpredictable system behavior. Join us next week for a deep dive into the missing piece in the AI stack: the Control Plane.

Building a self-service developer platform the thoughtful way
On-demand

Developers want a public cloud-like experience — even on-prem. In this insightful session, we uncovered how to automate full-stack Kubernetes deployments with a single Git commit. The entire webinar is now available on-demand.

Scaling real-time AI & ML workloads for performance and efficiency
On-demand

This insightful session on scaling real-time AI & ML workloads is now available on-demand. Learn where traditional infrastructure falls short, what modern AI/ML workloads demand, and how purpose-built systems enable predictable, efficient scale.

AI-powered K8s observability best practices in 2026
On-demand

As AI workloads move into production, the ephemeral nature of Kubernetes is becoming even more difficult to manage. Catch the replay to learn how to leverage AI-powered insights to ensure the health, performance, and security of your increasingly complex K8s environments.

New roadmap: Vibe Code with AI

"Vibe Coding" isn't just about writing prompts. It's about evolving from a manual coder to a technical conductor. This brand-new Vibe Coding Roadmap from our sister site, roadmap.sh, offers a comprehensive guide to building software in the AI-first era! It outlines how to leverage AI agents to handle the syntax while you focus on architecture, logic, and product vision.

Connect with The New Stack