In this month’s edition of the DirectorPlus newsletter, the Netlify CTO talks about how AI will enable the next generation of engineers to do things they haven’t dreamed of yet.
The future of AI-assisted development is so greenfield it’s neon-green, says Dana Lawson, CTO of Netlify. AI will enable the next generation of engineers to do things we haven’t dreamed of yet, she says. “It’s some science fiction shit.”
While AI shows promise, we’re still in the early days. Software development teams are just learning how to approach AI and integrate it into their daily workflows. Not everyone is on the same page tooling-wise, and there is still a cultural stigma around adopting large language models (LLMs). While the quality is steadily improving, it’s still dangerous to blindly push AI-generated code into production.
Lawson falls on the optimistic side. She sees the potential for generative AI to radically improve the developer experience (DX) for tomorrow’s technology teams, enhancing their lives rather than replacing their creativity. “If you want to future-proof your team and your stack, you need to embrace this technology,” she says. At Netlify, they’re walking the walk – replacing toilish tasks with AI automation.
Step 1: Establish a baseline
Machine learning has been around for a while, but things have been happening faster lately. As a result, this sudden shift has resulted in a hodgepodge of AI adoption across most organizations.
First and foremost, you need to understand where people are at with AI. To get a handle on things, Lawson surveyed the entire company, including both technical teams and non-developers, on what they were using. “Don’t just assume everyone is on the same page,” she says.
Step 2: Determine solvable DX problems
After getting a temperature check, it’s good to set intentions and reign in scope. You can’t buy every AI tool, says Lawson, so be selective in the acquisition process and narrow down what it needs to solve.
At Netlify, the overarching goal is to “eliminate roadblocks and friction to help developers do their dang job,” says Lawson. They’re committed to identifying and solving problems with DX, helping developers ship code more reliably and consistently.
Three examples of solvable DX problems
Take revisions and updates. Code migrations and updates can be gnarly and often introduce massive changes. Whether updating a CLI, API integration, re-writing feature flags, or cleaning up bad code paths, updates can require serious refactoring.
Streamlining SRE tasks
Netlify has partnered with Codemod and Codegen.sh which provides an AI tool to help with revisions and updates. It claims to be able to automatically find all the places that would break with an update. This is helping streamline SRE-like tasks that used to get a response like, “Aww man, am I on that job this week?” she says.
Coding with AI assistants
Netlify is a GitHub shop that allows its developers to adopt the Copilot coding assistant, primarily for “dumb code,” says Lawson. The premise is to remove friction from the developer journey and move from code to app as quickly as possible, freeing up developers to build value.
While adoption started slow, AI-generated code is slowly creeping into production at Netlify. “I had a pull request last week where 80% of it was written by AI,” she says. Although she notes it wasn’t feature code, it helped solve a problem.
Lawson wants to empower developers with the tools they need and encourage safe, smart use. “Nobody wants to be told what to do,” she says. “Let them solve the problems and give them the tools to do that.”
Making sense of codebases
Another area where AI has proven to be effective is in making sense of opaque codebases. The reality is that most documentation is a maze of forgotten Wikis, Notion docs, Google pages, spreadsheets, and more. “Not many places have good documentation management,” says Lawson. This lack of awareness can easily leave developers in the dark and stunt the experience.
This is where LLMs can help. Whether summarizing a ten-page product requirement document, or traversing hundreds of internal Wikis, AI can expose decisions around why code was written, says Lawson, setting engineers up for success early on. “How do we enable them even before their hands are on the keyboards?” she asks.
Step 3: Be aware of the risks
“At scale, there are always ghosts in the machine,” says Lawson. Of course, due to code quality and security concerns, it’s vital to put guardrails around AI-generated code. The GitHub ecosystem has helpful AI SecOps tools for this purpose, she says.
Also, Lawson notes that it’s good to test AI before you build it yourself. Netlify, being a DevOps platform for web development and deployment, intends to develop its own AI-driven features. Yet, it’s important to see it in practice first. “We have to be practitioners, so we know the problems other teams are facing,” she says.
AI usually needs a bit of nudging to get to the finish line. At the end of the day, you need to ensure AI improves efficiency and doesn’t cause additional rework to the point where it’s a productivity blocker. In short, the ROI has to be there.
Step 4: Overcome the stigma around AI
Culturally, Lawson notices three camps emerging around AI. First are the senior engineers who have been programming for decades, saying, “I can code better than LLMs” The second camp is made of divisive developers, saying, “My LLM is better than yours.” And the final group is junior developers, who may place too much faith in the outputs of AI-generated code.
Divisive opinions aren’t always great for team dynamics. But, of these camps, the greenfield developers pose the most concern. “They’re new and don’t have confidence in those coding patterns,” she says. “Can you call BS out of a model?”
Divisions like this could cause adverse outcomes, especially if folks are embarrassed to admit they are using AI. Lawson notes that everyone is on their own journeys, learning differently. To her, the best way to overcome stigma is to embrace modern toolsets and create a culture of humility.
DX for the long-term
“It’s our mission and responsibility as technological leaders to advance our teams with the skills and knowledge they know and give them that space,” says Lawson. “The developer persona is going to morph into more shared responsibilities.”
Tech trends change daily, and future-proofing your stack means future-proofing your team to ensure they can build and secure it. Although the industry shifted a bit away from the age of the “full stack developer,” AI has the potential to once more make developers more traversable across the stack.
Of course, many areas, like legal, security, and defining open-source AI, still need to play catch up. “It’s so green it’s neon-greenfield,” says Lawson. Nonetheless, embracing an AI-first attitude could become a matter of developer retention. No one is excited about working on a legacy stack with legacy tech. As such, if you want the best of the best, you have to embrace these tools, she says.
