Sam Saffron and the cost of a little waiting

Sam Saffron wanted to share a window. His computer offered a list of names, a screen tab he usually did not want, and a checkbox whose purpose seemed to require a separate negotiation. The machinery for having a conversation had become something to have a conversation about.

In November 2025, he described rebuilding that screen-sharing picker with AI assistance. There were blurry previews to fix and a crash to debug. Eventually, the dialog worked the way he wanted. A small irritation had become a small piece of software, made for one particular person.

The interesting part came afterward. Saffron did not think his useful personal fix automatically deserved to become somebody else’s maintenance job. Getting code to work on his machine and asking a project to take responsibility for it were different commitments. He understood enough to enjoy the result, and enough to recognize the distance still to travel.

That distinction runs through his career. Saffron, the Sydney-based co-founder and co-CEO of Discourse, has spent years making the machinery behind online conversation easier to examine and improve. His projects keep returning to a deceptively ordinary question: where does the time go?

A television, a forum, a job

The habit started well before Discourse. He received a ZX Spectrum at seven. Today, his interests away from the keyboard include rock climbing and fatherhood; the company biography also credits him with a steady supply of dad jokes. Anyone expecting the performance specialist to ration puns has chosen the wrong budget.

By 2010, he had built Media Browser, an open-source C# plugin for Windows Media Center. It let people browse their media through an interface designed to be seen from across the room. Supporting its users led him to build Community Tracker, a Ruby on Rails support tool influenced by Stack Overflow.

There is a pleasing circularity here. He used a community to support software, built software to support that community, and attracted the attention of people building another community. Stack Overflow introduced him as employee number eight in June 2010. Participation had turned into employment without ceasing to be participation.

Before that move, he had managed development for the Altiris platform performance team at Symantec and built profiling products there. The interest in measuring software was already a professional concern. At Stack Exchange, it would acquire an unusually public set of outlets.

Leave the workings where people can see them

One of those outlets was Stack Exchange Data Explorer. It gave people a way to run SQL queries against public data dumps, turning the accumulated activity of a question-and-answer site into something people could investigate themselves. A community could ask questions about its own questions.

In the 2012 update, Saffron described work with Tim Stone that added query histories, forking, attribution, and graphs. A shared query could improve without losing its permanent link. Those details sound modest until you imagine trying to learn from a useful example whose corrections have vanished into somebody else’s private copy.

The same update brought Data Explorer onto Dapper for data access and used MiniProfiler to inspect performance, including in production. These were tools Saffron had helped build at Stack Exchange. A piece of public software was also a working demonstration of other pieces of public software.

His departure in June 2012 was unusually candid. He wanted a bigger stake in what came next. “I want to be a founder or a mercenary,” he wrote. He also wanted to return to Ruby. The ambition was explicit, as was the attachment to leaving useful work behind for other people.

The chart catches the culprit

By March 2013, Discourse was becoming a practical setting for that combination of interests. Saffron was working with Jeff Atwood and Robin Ward on open-source discussion software. The public launch came that year; version 1.0 and the hosting service followed on August 26, 2014.

Around the launch period, he added flame graphs to Ruby MiniProfiler. Ruby 2.0 made collecting stack traces efficient enough to build a sampling profiler in Ruby itself. The result arranged sampled calls visually, so a developer could look for places where an unexpectedly large share of execution was accumulating.

When he aimed it at Discourse, mundane jobs came into view: constructing URL slugs, pluralizing words, looking up translations. In that particular front-page investigation, pluralization consumed eight percent of measured time. Grammar, it appeared, had submitted a rather ambitious invoice.

The appeal of this work lies in its specificity. An application can feel slow for any number of plausible reasons. A graph gives an argument something concrete to answer to. It also lets someone else examine the reasoning, rather than accepting a performance specialist’s intuition on trust.

A smaller bill buys a larger audience

Discourse gave performance work a purpose beyond tidy charts. In 2019, Saffron explained that memory use affected whether people could run the software on inexpensive servers. The difference between fitting on 512 MB and needing 1,024 MB mattered to adoption. Those were the constraints he was discussing then, rather than a statement of today’s system requirements.

He resisted making every line ugly in pursuit of theoretical speed. On code he found unnecessarily elaborate, his preference was blunt: “I prefer clear and dumb.” He would bypass an abstraction where a busy route justified it, while preserving readable Ruby elsewhere. Economy required judgment about where the expense actually was.

The phrase makes a useful companion to the graphs. A profiler can identify costly work; it cannot decide how much complexity a team should accept to eliminate it. Someone still has to live with the change. Software has a running cost and a reading cost, and Saffron’s choices acknowledge both.

“I prefer clear and dumb.”

Sam Saffron · 2019

Two CEOs, one long conversation

On February 1, 2023, Saffron and Sarah Hawk became Discourse’s co-CEOs. He had been CTO; she had been COO. Atwood moved to executive chairman. The arrangement placed technical and operational experience together at the top of a company organized around people talking to one another.

The business had reached $11.5 million in annual recurring revenue by the leadership announcement and had raised a $20 million Series A in 2021. These were company milestones under the preceding shared leadership, rather than a personal scorecard for either incoming CEO.

There was also a longer relationship underneath the titles. Atwood and Saffron had worked together at Stack Overflow before founding Discourse. Hawk had first encountered Discourse as a customer and community practitioner. The succession drew on people who already knew the product from different sides of the discussion.

For Saffron, leading the business did not end the public engineering work. His personal writing continued to tackle the practical details of building software. The subject grew to include AI, bringing a new abundance of code and a familiar shortage of time to understand it.

Cheap code still sends somebody a bill

His October 2025 essay arrived with an impatient title: “Your vibe coded slop PR is not welcome.” The complaint was about unfinished contributions presented for review. Producing a large patch had become cheap; checking whether it belonged in a maintained project still demanded experienced attention.

He proposed being explicit about what a contribution was ready for. An experimental branch could demonstrate an idea and invite discussion. A pull request meant the contributor had read the code, checked it, and was prepared to stand behind it. A working demonstration could be exciting while still carrying a great deal of unfinished work.

The position came from someone actively experimenting with agents. He described using his Discourse container tool, dv, to explore ideas in separate environments. Curiosity was welcome. Transferring the cost of that curiosity to an unsuspecting reviewer required more care.

There is a social courtesy buried in that engineering argument. Every contribution arrives in somebody else’s working day. A machine can produce text or code at a pace no maintainer can read. The person pressing submit remains responsible for deciding whether the result deserves another person’s attention.

Keep the door open

By April 2026, the discussion had reached the future of Discourse’s own source code. Saffron publicly committed to keeping it open. AI could help attackers inspect software, he argued, but it also gave defenders better ways to examine and repair a shared codebase.

He described a process in which suspected vulnerabilities had to be demonstrated with a failing test before reaching human review. The insistence on verification sounded familiar. Faster discovery would be useful when the team could establish what was real and act on it.

The company’s October 2026 plans extended that thinking into collaboration. Discourse’s proposed direction for 2027 brought people and AI agents into the same shared workspace, with discussions and decisions retained as context. It also renewed the commitment to community ownership and open source. These were announced intentions, with the work still ahead.

Saffron’s career gives those intentions some texture. A media library needed support. Public data needed a way to be queried. A slow page needed an explanation. A screen-sharing dialog needed a preview. The scale changes; the invitation to look closely survives.

He has made room for experiments without treating every experiment as finished work. That seems a useful temperament for the next stretch of software: interested enough to try the new tools, particular enough to inspect the result, and courteous enough to remember who has to read it next.