DevToolsDigest: Issue #388
Jeff Martens is CEO of Determinate Systems, where he works on making software development and delivery more reproducible, secure, and reliable with Nix. He's a two-time founder and led product teams at companies such as New Relic and Pagerduty, with a focus on the tools devs depend on to build, ship, and operate software at scale.
He believes that as AI makes software creation faster and more autonomous, and the production targets more consequential, the systems around us need to become more deterministic for the sake of reliability and security.
Subscribe to the DevToolsDigest
All you need is 3 minutes a week to stay ahead of the devtools space. Get the most relevant industry updates, insightful discussions, and top-notch resources delivered straight to your inbox.
Agents Expand Software Supply Chain Attack Surface
In his AI Council 2026 talk, Socket founder and CEO Feross Aboukhadijeh covers recent package compromises, vulnerability discovery, and a more automated security model.
Agents are starting to make dependency and tooling decisions at machine speed, while most supply chain controls were built assuming a human was in that loop. That assumption needs to change.
The Next Version of the Model Context Protocol is Here
The 2026-07-28 MCP specification is out, bringing a stateless protocol core, Multi Round-Trip Requests, header-based routing, cacheable list results, authorization hardening, a formal extensions framework, and updated Tier 1 SDKs.
MCP is growing up into real infrastructure, and the boring stuff is the point: stronger authorization, explicit metadata, statelessness, predictable ordering, and a real deprecation policy. These are exactly the kinds of constraints agentic systems need as they move into production.
The AI-Era Software Assembly Line
To understand how AI-era software assembly differs, Sonatype Research Labs examined four years of software assembly data to understand how software risk has changed since the beginning of the AI era.
Even as teams get faster at fixing vulnerabilities, Sonatype found that critical and high-risk exposure per application has grown more than 4x thanks to AI. Supply chain security needs to move earlier, to the moment developers and agents choose which dependencies become part of the build.
AgentBaiting: Fake AI Skills Deliver Malware at Scale
Inside the 7,600-repository FakeGit operation that brought SmartLoader into the AI capability supply chain, using GitHub repositories, public AI registries, and agent-readable instructions to create a new enterprise attack surface.
What stands out is that the same agent could reject a malicious repository in one run and recommend it in another. That’s a great example of why discovery and installation decisions need hard policy outside the model.
What Is a Dependency Firewall?
A dependency firewall can protect developer workstations, build environments, CI/CD pipelines, and AI coding agents by blocking packages that appear malicious, suspicious, or inconsistent with organizational policy.
If agents are going to choose dependencies, controls need to move in front of installation. Decide what software is allowed into the environment before an install script gets the opportunity to execute.
Deterministic AI Is an Architecture Problem
Ask Claude Code to stand up a Virtual Private Cloud (VPC) and it will. Ask again tomorrow and it will again, but the result will look nothing like the first one. That gap is the whole problem with moving agents from demo to production.
A clean articulation of the idea behind this issue: don’t try to make the model deterministic, make the architecture around it deterministic. Let the model provide judgment, then let tests and validation decide what is acceptable.
AvioNix: Infrastructure as Code for Linux in Avionics
As the community continues exploring Linux for safety-critical aviation applications, AvioNix demonstrates how Infrastructure as Code can help improve reproducibility, traceability, and collaboration across open source development.
When reproducibility stops being a developer-convenience problem and becomes a property of safety-critical infrastructure. AvioNix shows how Nix can bring explicit inputs, isolated builds, and traceability to Linux systems where “works on my machine” really isn’t good enough.
Edge AI Stack and the Fleet
Edge AI has moved inference off the cloud and onto the device, and the accelerator silicon is no longer the limiting factor. But an accelerator is inert until a software platform stands it up.
AI is driving a boom in robotics, making the delivery and security problem much harder: models, runtimes, drivers, firmware, kernels, and hardware all have to move together across a fleet.