Builderloo
All stories
Builderloo story

When Building Gets Cheap, Owning Is What's Left

Torrin Leonard · November 20, 2026

Abdelrahman Awad of Sentry told Waterloo.dev Fest what AI has changed for people who maintain open source, and the closing panel took on whether coding is dead

Abdelrahman Awad on stage at SPUR with his introduction slide behind him
Abdelrahman Awad on stage at SPUR with his introduction slide behind him

On November 7 I spent the day at Waterloo.dev Fest at SPUR, which Waterloo.dev ran with the Google Developer Group in Waterloo. Jim Balsillie gave the morning keynote, and I wrote that one up on its own.

The talk right after it was about open source and AI, from Abdelrahman Awad, who has had strangers sending him code for ten years and now gets a lot more of it written by AI.

Ten years of nights and weekends

Abdelrahman is a senior engineer at Sentry who lives in Toronto. He made VeeValidate and Formwerk, two open source libraries that web developers use to build forms.

Abdelrahman started VeeValidate ten years ago as a late night fix for his own problem. He said he needed a form to check email addresses and links, and nothing out there worked the way he wanted, so he wrote it, used it in his own projects, and then shared it to see what other people thought. His slide showed what it grew into, and it's a lot for something he did on the side.

4.9Mdownloads a month
11kGitHub stars
411people helped build it
96%of 3,578 bug reports and questions answered

Abdelrahman did almost all of that on evenings and weekends. Early on he built every feature anyone suggested, the library got too big and did too much, and when he cut it back down he lost users. He said that is how he learned to care about the people using a project and still be the one who decides what goes into it. I liked that he put his own mistakes up on stage.

Abdelrahman also said the fun part of the project ended years ago and most of the work now is keeping it running for other people. A bug report takes a minute to file and can take him an evening to figure out, and a fix for one person can break something for someone else.

Getting paid to do it

Abdelrahman walked through the ways open source gets paid for. Sponsorships through GitHub Sponsors, Patreon or Ko-fi mostly reach the famous projects, and in his view a smaller project, one without a marketing push, or one run from another country usually misses out. Companies and governments also give grants, and he got one from Google once, but they were never much of his income.

The path Abdelrahman chose was getting hired to do it. Sentry pays him to work on open source full time, and he said most of the code he has written in his year there is in other people's projects, because when a tool Sentry depends on has a bug, his job is to go and fix it there. He also told us about the Open Source Pledge, which Sentry started. Companies that join promise to pay open source maintainers every year, and he said 37 companies are in it now and Sentry raised its own budget to $1 million this year.

Abdelrahman was at the Node.js Collaborator Summit this year. He said the meetings where the future of Node.js gets decided are open to anyone who joins and has something useful to say.

Open source and the future, definitely defined by those who show up.
Abdelrahman Awad

What AI changed for maintainers

Abdelrahman called himself an AI skeptic who uses it every day and pays a lot for it, and he gave both sides. He said AI does the boring chores like upgrading the other libraries a project depends on, it can reproduce a bug in minutes instead of an evening, and he has shipped things in weeks that would have taken him a year.

The bad side, according to Abdelrahman, is a lot more low effort reports and code. His example was someone who asks their agent to find a bug in a popular project and files whatever it finds without ever having run into the bug. He pointed out that he could ask his own agent to do the same thing, so a report like that doesn't add anything.

Abdelrahman showed numbers from a talk Rafael Gonzaga gave on Node.js security at the Node.js Collaborator Summit 2026.

733people sent security reports to Node.js
3people did half of the work of reviewing them
40 to 50invalid reports a month in 2026, up from a handful

Node's answer, from the same slide, was to make those reports cheap to answer too, with AI doing the first pass and a person making the call. Abdelrahman said some projects went further. Ghostty, a terminal app he uses himself, now turns away what it calls drive by AI contributions, and he named others that stopped paying people for finding security bugs or stopped taking code from new contributors.

Abdelrahman's explanation is that writing code has become nearly free and reviewing it has not. He said that when people wrote code by hand they checked it as they went, so review was the last step. Now a maintainer can get a thousand line pull request that nobody has read and has to check all of it as if they wrote it themselves.

Abdelrahman's own team at Sentry hit this. He said their open pull requests usually sit around 30, at one point they reached 120, and they were around 87 when he spoke. Good contributors end up waiting in the same line as everyone else, so his team is going to try a few changes. Before you send code, you open an issue and talk to them, and a maintainer has to give you the go ahead. Work that reads like a person cared about it gets looked at first, because, in his words, "some care went into this, so we should give some care back."

Owning what you ship

Abdelrahman's slide that reads "When building gets cheap, owning is what's left."

Abdelrahman's slide that reads "When building gets cheap, owning is what's left."

Abdelrahman said that at Sentry there is no rule on how much AI you use, but you own every line you ship. If something breaks, "my AI did that" is not an answer you can give your manager, so many people on his team read their own code line by line before they ask anyone else to review it.

Then Abdelrahman asked those of us in the room about the polished pull request someone sends to a project after an AI wrote it. Could the maintainers make the same thing with their own tools? Could the person sitting next to you? He said yes to both.

We don't need your tokens. Everyone has tokens.
Abdelrahman Awad

What a project needs from you, Abdelrahman said, is your insight into why something is a problem and how it affected you, your judgment on how it should be fixed, and the follow through to get it merged. His advice for anyone who wants to start was really practical, and a student could do it next week. Pick a project you actually use, look for issues labelled "good first issue", and ask to own one before you write anything. Filing a good bug report counts, and so does fixing documentation you found confusing or translating a project into a language you know.

Staying in control

In the questions, someone asked how his team handles code review now that so many people at work are writing code with AI. Abdelrahman said they use an AI reviewer, take what it says as a suggestion, and have another agent check its claims. He said it catches things they would never have thought of, and lately it also flags a lot that turns out to be nothing.

Abdelrahman's bigger answer was about staying in control. He told us that a few months ago he lost track of what was going on in a set of his own pull requests, with review comments coming in and the agent changing things faster than he could follow. He stopped, read everything again, and made a command that cuts an agent's long answer down to three lines so he can follow it.

You steer the AI. Don't let it steer you. Don't become a meat proxy.
Abdelrahman Awad

By a meat proxy, Abdelrahman meant the person who pastes a coworker's question into an AI and pastes its answer back without thinking about it.

The closing panel, on whether coding is dead

The day ended with a panel called "Who owns, governs and builds our infrastructure?" Sanjay Samuel of Waterloo.dev hosted Bianca Wylie of Time & Space Advisory, Brett Gillett of Curious Orbit and Vlad Gheorghiu, the CEO of softwareQ, a Waterloo company that makes software for quantum computers. I only caught the last 15 minutes, when someone in the audience asked whether coding will be dead in a couple of years.

Sanjay Samuel, Vlad Gheorghiu, Brett Gillett and Bianca Wylie on stage at SPUR

Vlad said coding the way we did it five or ten years ago is a little bit obsolete. He has AI write the code he used to look up and copy, writes the tests himself, and makes sure he understands the result. The modern coder, he said, spends more time designing how the software fits together and deciding what to ask the AI for.

Brett said he hasn't written a line of code this year, but he reads far more than he used to. "Read what it writes," he said, and question it when it doesn't look right.

Bianca, who doesn't write code, said that when she writes something herself it "starts to live in my body", and reading a summary is not the same. When you have to defend something, she said, you can tell who knows a thing and who only read a synthesis of it.

Do you maintain something here?

Thank you to Waterloo.dev and the Google Developer Group for putting on a full day, and to SPUR for hosting it. If you maintain an open source project in Waterloo Region, or you have started contributing to one, I'd love to hear how AI has changed it for you. Send me a note through the contact page.

In this story

People

Torrin Leonard

@torrin · Founder & Engineer @ Builderloo

Author

Software Engineer and Founder at Builderloo. Building tools with ai for Canadians and local communities.

Startups

softwareQ

Designing quantum software

Quantum ComputingDeep Tech & Scientific R&DB2B

11-50 employees · since 2017

Hubs

SPUR Innovation Centre

A frontier-tech venture studio and AI data-centre operator on the former BlackBerry campus in Waterloo, with office hours and events for founders.

Waterloo, former BlackBerry campus

Sign in