EXCLUSIVE An internal memo dispatched by senior execs at Red Hat suggests the software biz is starting to push AI tooling within its Global Engineering department. RHEL may be about to get some Windows 11-style "improvements."

The Register has seen the missive signed by "Chris & Ashesh" – Chris Wright, Chief Technology Officer and Senior Vice President of Global Engineering, and Ashesh Badani, Senior Vice President and Chief Product Officer.

It carries the heading "Engineering that's evolved and amplified for the AI era," and for any AI skeptics in the developer teams at Red Hat, the tone of the email may raise alarm bells. The times are changing, it states.

"Our roles: All Global Engineering roles will evolve. The focus will shift from 'AI as a tool used on occasion' to 'AI automation as a way to scale the delivery of value to customers.' Our associates' skillsets will grow as they become proficient in these tools."

This sounds to us like Global Engineering's management is determined not only to encourage the use of AI tooling, but to require that team members learn how to use it. This will be an uncomfortable time for those who prefer not to use such tools.

This pattern mirrors that at some other enterprise vendors. In November, The Reg reported on a Microsoft exec who found it "mindblowing" that some people still do not believe in the powers of large language models. This goes all the way to the top of the company. At the start of the year, CEO Satya Nadella stated that it was past the phase of "discovery" and into "the phase of widespread diffusion."

Red Hat did not immediately respond to our request for comment; we'll update this story if they do.

A detailed rundown

The Red Hat memo continues to discuss "our processes," saying "the Global Engineering software and product development lifecycle is going to transform. AI will allow us to deliver on challenging use cases and lifecycles that we say 'no' to today. And we'll continue to learn about and adopt industry best practices – embracing what works for Red Hat, and leaving behind what doesn't."

The plan is that "what we do today in open source communities is what we'll continue to do tomorrow." It hopes to "lead by example," influencing developers to be "more AI-friendly and sharing our success stories broadly." This involves building frameworks, establishing best practices, and defining standards – to shoehorn AI into the open source development lifecycle.

The language the execs use paints the sky as very blue, but any techies with suspicions about the clouds of hype around AI may have reservations about what looks like a revolution forming at Red Hat.

Wright and Badani reckon that since discussing the "opportunity to build the next version of Red Hat" in a world "changed by AI" in fall 2025, things have only accelerated. A visible external sign of this was Wright's public blog about "supercharging" AI-assisted development from September last year. We suspect that this perceived acceleration may refer to the claimed shift in coding assistants' abilities that we mentioned in the closing paragraphs of this late-February story. We have seen similar claims in multiple places in the last couple of months.

It continues: "Our competitors are not just 'using AI' – they are reorganizing their entire workflows around agentic systems to ship at a velocity we must match to maintain our leadership." As one Register sponsored feature notes, agentic AI is indeed the buzzword of the moment.

We feel it's important to clarify which "competitors" Wright and Badani mean. The company's perspective is that its rivals are major enterprise software vendors: Microsoft, Oracle, Broadcom, and so on. Red Hat sees itself as the upstream developer of the most important Linux projects and, as such, all other Linux companies are downstream of Red Hat, and thus not as significant to the company.

The next few paragraphs of the memo are oddly repetitive. First, it says: "To lead in this era, we must evolve our operating model. The gap we face today isn't just technical – it's organizational. Today, we are beginning our transition to an Agentic Software Development Lifecycle (SDLC) to transform how Global Engineering and Products deliver."

It then continues to largely repeat the same point, complete with explaining the same initialism again: "We are transitioning to an Agentic Software Development Lifecycle (SDLC). AI will be the operating model we run on. This is not about 'speeding up old processes'; it is about a world-class, agent-first development model fundamentally increasing the volume and quality of what we ship."

We feel obliged to note that such repetition can be a sign of text that was generated by an LLM, leading us to suspect that the authors may have used such a tool. As we reported from the CentOS Connect conference last month, we sat behind a Red Hatter doing just that in a message to an email distribution list.

"In this model, our teams provide critical human oversight focusing on architecture, judgment, strategy, and complex content. We are the orchestrator with agents as the execution engine," it adds.

This is a nod to the use of the term "orchestrator" in the context of container orchestration tools, of which Kubernetes is by far the most widely used. As we noted in the bootnote to our recent story on the Ladybird browser, noted coding commentator Steve Yegge has this year become evangelistic about running entire automatically managed swarms of programming bots. Since the articles we linked to, he has only become more and even more enthusiastic.

On it goes: "Outcomes, not tools: We define goals by speed, cost, quality, capacity – not specific models. If a better agent framework emerges, we adopt it. Redefine our workflows through agents: Every process and workflow must have a way to onboard agents to drive execution. Individuals provide context, shape with feedback, and oversee agents."

The metric is the workflow level, not the activity level, the pair continue, where teams are measured by whether workflows are hitting targets, including cycle time, defect rate, throughput, and resolution time. We wonder whether a management team so zealous to set targets about workflow are familiar with the ramifications of Goodhart's Law.

"'All-in' product scope: We aren't looking for a single scrum team to experiment in a vacuum. To avoid bottlenecks, we will move entire products or sub-products to this model simultaneously." Scrum is a reference to the Agile development model, about which The Register's Rupert Goodwins expressed reservations back in 2024.

While "our commitment to open source and upstream is not changing," the execs admit that "product and project development processes may diverge initially as we focus on how we build and deliver our products." Red Hat will try to "influence community development processes such that our processes can converge over time." It is possible this means the company might attempt to get external development communities to adopt similar practices – and that the authors anticipate significant resistance and even pushback from some communities.

"Given the broad spectrum of projects we work with, this will happen quickly in some areas and slower in others... We will transform the entire lifecycle, from planning (with BU participation) through designing, coding, code review, testing (unit and E2E), quality control, security, building, signing, documentation, support, and sustaining."

We suspect that they're right on this point and the preceding one: yes, both adoption and the results will definitely be uneven. It could be, though, that the tools will turn out to be much more useful for code review and testing than they are for writing it or documenting it, for instance. Notably, Greg Kroah-Hartman, the manager of the Linux stable series of kernels, says it's recently got much better at bug hunting and even fixing.

For any doubters out there, Wright and Badani say this is not about their talented coders becoming "prompt engineers who rubber-stamp output. It's about delivering features, security, and quality at the pace the market demands." As Bruce Schneier noted nearly a decade ago: "We know the market does not pay for quality software. The adage is good, fast, and cheap: pick any two. The market has picked fast and cheap at the expense of good. Pretty much everywhere, software doesn't work very well." It could be he will turn out to be even more correct than he realized in 2017.

Where is this happening? "Some teams in Global Engineering are already undertaking an agentic-first approach. We asked these teams to accelerate their transformation, and many already have. For example, Kevin Myers' Ansible engineering team is evolving both what they build (bringing agentic capabilities into AAP) and how they build it (adopting AI-driven development practices). Kevin's team didn't wait to make this change – they couldn't. We've given them three months to make this happen, and other teams are following the same path."

The speed that other teams move to this model is governed by "market needs" and organizational leaders will outline guidance to "prevent redundant experimentation," the message says. "We are currently defining leading quality indicators, which include code coverage and PR cycle time (among others), and how we're accelerating development – through feature velocity or time to market – that will show us how this transition is taking hold. This data-driven view will allow us to adjust quickly and scale what is working across the organization."

Urgent! Urgent!

As the last effects of the Kool-Aid passed, the pair signed off. "The opportunity is real, and the urgency is high. We have the talent and the intent; now we need the collective discipline to standardize and scale."

Clearly, there is a strong sense of urgency here. Such haste is something one associates with smaller companies and startups, but we have seen large companies pivot with remarkable speed before now. The prime example certainly seems to be on the minds of Wright and Badani: Microsoft. The ancestor of all modern versions of Windows, Windows NT 3.1, shipped in 1993 without any form of web browser; two years later, so did the first release of Windows 95. Internet Explorer 1.0 was only found in the paid add-on Microsoft Plus!. Within a year, Microsoft was industriously paddling the web boat. By summer 1996, both Windows 95 OSR1 and Windows NT 4 came with Internet Explorer 2.0 included, and NT 4 also offered Internet Information Server – all free of additional charge.

We found the use of the term "discipline" toward the end of the memo rather chilling, but it may prove necessary. Red Hat is not the only employer finding that it has to mandate employees to adopt AI tooling. In February, The Register reported on such a move at Accenture, and in March at PwC. This is despite 2024 research showing that workers say it doesn't help, and multiple studies showing companies and their execs struggling to show a significant return on investment.

It will be interesting to see if this "data-driven view" will be flexible enough to reverse the AI push if the performance indicators show no clear benefits. From the evangelistic tone of this memo, we suspect not. ®