How to train your developers

Welcome to Runtime! Today: Microsoft's Brendan Burns thinks there's a path for the junior software developer in the AI age, OpenAI takes another step toward the enterprise, and the latest enterprise moves.

Please forward this email to a friend or colleague! If it was forwarded to you, sign up here to get Runtime each week, and if you value independent enterprise tech journalism, click the button below and become a Runtime supporter today.

The golden path

As AI software development tools start to gain traction inside mainstream tech organizations, it's becoming clear that a fair amount of the traditional methods for evaluating and training those developers will need to evolve. There are some upsides to that shift, which might put the final nail in the coffin of the "how many lines of code did you write today?" standard, but it has also raised questions about how junior developers will learn the skills it takes to become a well-rounded software engineer if they aren't toiling in the coding mines.

"We're moving from a world where you learned sort of implicitly and over time to a world where you need to be explicitly taught," Microsoft's Brendan Burns told Runtime in a recent interview. Burns, who currently leads a 1,400-person engineering team at Microsoft working on Azure, agreed that the profession is going through real change and developers will need to be judged on different skills going forward.

• "When I [got] my computer science degree, I spent one semester on software engineering; I spent a lot of time learning how to write code and a little bit of time thinking about, how do you do software engineering?" he said.
• AI tools change that equation, where everyone is basically on the same level when it comes to actual coding and skills like project design become much more important.
• As Amazon showed the world last week, code review is becoming a much bigger bottleneck in the software engineering process as AI tools crank out increasing amounts of code, but on traditional software teams code review is done by senior staff.
• There's no reason why junior engineers can't be trained to do code review, Burns said: "We need to figure out how to have new grad engineers also be code reviewers, and that involves teaching them things, and that involves figuring out what a curriculum or an educational system is for code review."

But AI tools are creating a new level of abstraction that could one day reduce the need for code review at all by making a lot of code "transient," almost a throwaway part of the software-development process that engineers don't really need to worry about, Burns said.

• When he first started developing software in the late 1990s, Burns worked with engineers who had been through the transition from machine-level languages like Assembly to higher-level languages like C and C++ that were easier to use but needed compilers to translate that work so it could be read by the computer.
• The first few generations of C compilers weren't great, Burns said, which meant that those engineers frequently had to wade into the translated code to fix errors.
• But those compilers got much, much better over time, which meant that nobody had to read or even understand the machine code generated by the compiler; it just worked, and was therefore "transient."
• "There is a point out there, potentially, where you have a good enough spec and you have a good enough set of tests that you run, that the code generated by the AI doesn't matter in the same way that the code generated by the compiler doesn't matter; as long as the tests pass, as long as it meets the spec, it is operating correctly." Burns said.

Burns, who was one of the main creators of Kubernetes at Google before joining Microsoft in 2017, understands the angst that professional developers feel right now as something they once saw as a craft enters its mass-production era. But just as Microsoft's Jay Parikh told Runtime last year, AI tools have enormous potential to take on the tasks that developers really don't want to do, and allow them to focus on the work that matters.

• "Something I don't think we talk about as much, and I wish we talked a little bit more about with AI coding, is there's kind of this implicit assumption that the AI is writing code that a human would have otherwise written," Burns said. "But I think one of the things that we're finding is actually, like, the AI is actually writing code that nobody was going to write."

Astral's week

Faced with the reality that it needs to generate an actual ton of revenue over the next few years to pay for its infrastructure spending commitments, OpenAI's pivot to the enterprise is well underway. On Thursday it announced that it had acquired Astral, which builds developer tools for Python users, for an undisclosed amount.

Python is the lingua franca of the AI world, and Astral's tools help Python developers manage dependencies and check for errors in their projects. Astral's developers will join OpenAI's Codex team as it tries to chip away at the lead Anthropic has carved out in the enterprise world with Claude Code.

"By bringing Astral’s tooling and engineering expertise to OpenAI, we will accelerate our work on Codex and expand what AI can do across the software development lifecycle," OpenAI said in a blog post. It also said that it would keep Astral's tools open source after the deal closes.

Enterprise moves

Saahil Jain is the new chief technology officer of You.com, a promotion from his previous role as head of AI after co-founder and former CTO Bryan McCann left to join Anthropic.

Alfredo Hickman is the new chief information security officer at Kai, joining the security company after serving in security leadership roles at Obsidian and Rackspace.

Tim O'Neil is the new chief revenue officer at Matillon, joining the data automation company after sales leadership roles at Alation and ThoughtSpot.

The Runtime roundup

An in-house OpenClaw-like agent tool went bonkers at Meta recently and exposed sensitive corporate data internally to employees who didn't have permission to see it, according to The Information.

SAP's five-year plan to move customers of its old on-premises ERP tools to the cloud is well behind schedule, according to The Register.

Sometimes the headline says it all: "Federal Cyber Experts Thought Microsoft’s Cloud Was 'a Pile of Shit.' They Approved It Anyway," ProPublica reported.

Thanks for reading — see you Saturday!