State of AI Engineering

What The State of AI Engineering Report Reveals About AI in Production

AI engineering has moved past experimentation, but are your observability practices keeping up? Datadog analyzed LLM telemetry from 1,000+ customers to reveal what's actually happening at scale: model adoption shifts, hidden token costs, and what the rise of agentic frameworks means for reliability. Download the report to benchmark your AI stack against production reality.

Typescript breaks free from Node.js - Vercel’s new scriptc compiler turns TypeScript directly into native executables, removing the need for Node.js or V8 at runtime. It aims to improve startup times and reduce memory usage, making TypeScript more practical for scripts and command-line tools. Read more.

Qualcomm & Linux married- Qualcomm is bringing Linux support to Snapdragon X2, with early work focused on getting key hardware components running on the platform. The effort could make Snapdragon X2 laptops more useful for Linux developers and ARM-based workloads. An early developer preview is available, with broader Ubuntu support planned as the platform matures. Read more.

AI Killed the MVP. What Comes Next?

For years, the Minimum Viable Product, or MVP, was a central idea in startup building. The goal was simple: build the smallest useful product, put it in front of customers, and learn from what happens. But AI is changing how quickly software can be built. Today, a team can use AI coding tools to create a working product in hours, making the existence of a functional MVP much less meaningful as proof that a startup has learned something. Steve Blank argues that this shift requires a rethink of what an MVP actually tells us.

Why the MVP Is Changing

The original Lean LaunchPad approach treated an MVP as more than a prototype. It could demonstrate technical competence, help teams test customer problems, and provide evidence about product-market fit and the wider business model. Teams were expected to build, test assumptions, talk to customers, and use evidence to decide what to change next.

AI changes this equation because building the product is no longer necessarily the hardest part. With AI-assisted development, teams can quickly turn an idea into a working application. That means a polished-looking product created on day one may tell us very little about whether customers actually want it or whether the founders understand the problem they are solving.

Blank therefore proposes calling these early products Initial Untested Products, or IUPs, instead of MVPs. The name reflects the idea that a working product is not automatically evidence that its underlying assumptions have been tested.

Building Is No Longer the Main Test

The change does not mean AI should be removed from entrepreneurship. In fact, Blank argues that students should use AI extensively because it is now part of the environment in which modern companies are built. The important change is understanding what the resulting product actually proves.

A working application can now be produced quickly, but questions around customers, pricing, distribution, business models, and market demand still require evidence. The Lean LaunchPad method was built around testing these assumptions through customer development, experiments, interviews, and repeated learning rather than simply producing a product.

The Bigger Change

Blank's conclusion goes beyond renaming the MVP. He argues that AI has changed several conditions around startups, including how products are built, the cost and speed of development, venture capital, and how customers evaluate and adopt products.

The result is a subtle but important shift: when software becomes cheap and fast to build, the product itself becomes weaker evidence. The harder questions move elsewhere. Does the problem actually exist? Will customers pay for the solution? Does the business model work? Can the team explain why it built the product this way?

AI may have made the MVP easier to create, but that also makes customer evidence and validated learning more important, not less. The product can be built in hours. Finding out whether anyone actually needs it is still the real test.

Files can leak secrets! - Security researchers found that file-notification systems in Windows, Linux, Android, and macOS can reveal user activity without exposing the actual file contents. The leaked signals can potentially expose patterns such as keystrokes, website visits, file access, and authentication activity. Read more.

GitLab’s secret bug - A GitLab issue-by-email address can act like a credential, allowing anyone who gets it to create commits or trigger CI jobs using the account’s permissions. The address does not expire, making accidental exposure especially risky for developers and maintainers. The finding highlights why treating secret-looking email addresses as security controls can create problems. Read more.

Buzz of the Week!

Epistemic Containment

Epistemic containment is a niche concept for keeping uncertainty inside a controlled part of a software or AI system instead of letting it spread through the entire workflow. Rather than forcing an uncertain prediction into a hard yes-or-no decision, the system can carry its probability forward until more evidence is available. This matters when one uncertain output could trigger multiple downstream actions. For example, a fraud system could pass a probability score to another service instead of immediately blocking an account. The next component can then decide how much uncertainty it is willing to accept. The goal is not to remove uncertainty, but to prevent uncertain information from quietly turning into confident system behavior.

Things that launched. Things that went viral. Things you'll pretend to try.

cilium

cilium is an eBPF-powered networking, security, and observability platform for Kubernetes.

kubevious

kubevious is a visual tool for understanding Kubernetes clusters and spotting configuration problems.

confest

confest is Tests configuration files using policy-as-code. Particularly useful for Kubernetes, Terraform, Docker, and CI configurations.

Build Braincells, Not Just Features

This weekend’s read: The Republic of Slop.

This week’s watch: Why Did Human Progress Suddenly Explode After 300,000 Years of Slow Progress?

Meanwhile…