Matt Nigh: "You can’t force AI from the top down – it needs energy from the ground up"

Matt Nigh is the Program Manager Director of AI for Everyone at GitHub. This initiative supports the large-scale adoption of AI at GitHub, a widely adopted developer platform that is used by more than 150 million developers to build, scale, and deliver software. As part of his role, Matt has designed an “AI Playbook” that explains how GitHub created an operating model to support the rapid adoption of AI across the whole company.

Not too long ago, we saw a wave of CEO memos encouraging employees to use AI. However, it seems that these memos mostly created anxiety rather than inspiration. At GitHub, you’ve taken a different path. Could you explain your approach?

At GitHub, we believe that AI creates immense value for employees and teams when applied correctly. But AI isn’t something you can just “slap on” a process and walk away from. You need a clear understanding of why it’s valuable to each individual and their specific use cases.

That’s why we took a grassroots approach. It allowed us to best understand how AI can provide value across the wide variety of teams and employees that work at GitHub. Using AI in daily work is a fundamental change to how you work. It’s both a technological and cultural shift. You can’t force it from the top down – it needs energy from the ground up. At the same time, top-down support is essential for providing resources, investment, and time to learn.

Top-down support is important, but the real momentum comes from employees themselves, especially from internal champions who create space for people to share what they are learning

That balance – top-down and bottom-up – seems central. You’ve codified this in a “playbook.” Could you talk us through its pillars?

Sure; our model consists of eight pillars, including advocates, the right tooling, communities of practice, and clear policies and guardrails. If I were to distill it into three core elements, I would say the most important ones are top-down support; bottom-up support, specifically through AI advocates and events; and data so that you can understand your progress and adapt your programme.

The top-down piece is about getting leaders on board, securing investments into your tooling, creating mechanisms for your team to learn in their day to day work and setting clear and simple rules and expectations for the employees.

But the real momentum comes from employees themselves, especially from internal champions who create space for people to share what they are learning, offering practical training and creating a culture where people can constantly share what they are learning. This is where most of the impact comes from, but this couldn’t be successful without top-down support.

The last piece is where the data comes in to keep the whole thing honest. We have quarterly goals that focus on AI adoption and what we call AI fluency or – in other words – how effectively employees use AI tools.

Let’s dive into the metrics you just described. Metrics often define how new technologies are deployed, so how do you approach finding the right metrics?

We published a whole GitHub playbook on the subject, the “Engineering System Success Playbook”. This playbook looks at metrics in engineering, including AI-related metrics, in detail. But with regard to our AI program, I think about metrics in two ways: How to measure AI adoption and fluency – are people using the tools and are they skilled at it? And secondly, how to measure the benefits, for example changes in productivity, cost savings and so on.

For adoption and fluency, we measure active users, usage habits, use cases and employee satisfaction or whether people like the tools they are using. For productivity, we use time saved, a common standard in our industry, and revenue per employee. For engineering specifically, we track things like pull requests, code completions, and apply existing developer metrics to AI tooling.

When AI is used to offload maintenance work, teams can start to experiment aggressively

A recent MIT study showed that 95% of corporate AI pilots fail, often because they are top-down initiatives. Does GitHub’s more bottom-up approach explain your success?

In my own experience, I have seen nearly every team at GitHub gain value from AI, especially with GitHub Copilot. Engineers get immediate feedback loops. Product managers and designers, who may not be deeply technical, use tools like GitHub Spark to build software prototypes by just using natural language. And our product marketing teams use a specialised chat to draft briefs and so on. We see faster completion times, more space for innovation and experimentation, and quicker time-to-market.

Beyond individual productivity gains, do you see system-level changes in how GitHub operates?

Absolutely. Across the industry, people are getting good enough with AI to offload maintenance work. That frees teams to experiment aggressively. That would not be possible without them being able to delegate some of the work that they already had to AI.

In our team, we even created a policy encouraging people to set aside two weeks for experimentation every quarter. Personally, I built an internal website for learning and development for AI – I wouldn’t have been able to do that pre-AI in such a short time as only one person. That’s a very big difference and shows how companies shift with AI.

AI will lead to a surge within software development; in our industry, when new technology created productivity gains, demand for developers always increased

Let’s talk about skills. What capabilities matter most in this new world of work?

I think of AI as a partner, not a replacement. AI agents aren’t here to replace us, they are here to empower us. GitHub Copilot can handle some pieces of my work, for example maintaining my code base and that frees me up to work on other things that are more interesting like designing a complex system, thinking critically about the code not just from a technical perspective but from the perspective of a realistic business with all its constraints, and solving problems creatively.

As a result, professions of course change and need to adapt. Some skills in software development will become more important, for example architecture, strategy, and design. And some new skills will be introduced such as how to manage a team of AI agents or reviewing large volumes of AI-generated code. Some other skills, like maintaining simple codebases or creating proof-of-concepts, may lose value because AI can handle them.

Above all, I believe that AI lowers the barrier to entry. In the past, coding for example required a good knowledge of English. AI has changed that. So, I think AI will lead to a surge within software development. In our industry, when new technology created productivity gains, demand for developers always increased.

Sustainable AI adoption needs a bottom-up approach: Individual teams are best placed to decide how to adopt AI. But support from leadership top-down support for providing resources, investment, and time to learn equally essential.

Measure what matters: Some measures, such as AI adoption and fluency (are people using AI and are they skilled at it?) may be universal, others might be more specific to what the team does.

AI lowers the barrier to entry: Some skills may lose value because AI can handle them while others will become more important. Overall, however, AI will lower the barrier to entry for many professions.

Season 1: AI and the Labor Market | Episode 1: The Future of Work, In Progress | Episode 2: Carl Benedikt Frey: “Professionals are not prepared for the coming changes” | Episode 3: Jonas Andrulis: “Digitize the state! That’s the foundation we all stand on” | Episode 4: Cindy Richter: “AI creates roles that didn’t even exist before”