Brian O’Kelley and the trouble with a working machine

Brian O’Kelley turned up for the interview in a suit. This was his first error. The person meeting him looked like a dot-com executive; O’Kelley looked as though he had come to discuss the mortgage. The second surprise was the job. Right Media needed somebody to build a website and automate a few chores. He began explaining his experience with enterprise software and distributed systems. His interviewer suggested he speak to the technology company sharing the office instead.

In O’Kelley’s telling, that awkward encounter led to consulting work at Poindexter Systems, and eventually back to Right Media. It is an appealingly untidy entrance for someone whose career would become bound up with organizing the internet’s advertising machinery. Nobody in that room was commissioning an industry. They were trying to get a small business through its working day.

More than two decades later, the same engineer is proposing another rearrangement. He co-founded AppNexus, sold it to AT&T, left advertising for physical supply chains, and returned with Scope3 to measure the environmental cost of digital ones. In September 2026, Scope3 became Apostra, with AI agents at the center of its business. Each chapter raises a familiar question: what, exactly, should the software be doing?

The specification was perfect. The product wasn’t.

O’Kelley grew up in Eugene, Oregon, played basketball at South Eugene High School and graduated from Princeton in 1999 with a B.S.E. in computer science. His LinkedIn education entry includes junior varsity basketball. The résumé offers two kinds of teamwork, although only one usually makes it into an advertising conference introduction.

At Right Media, the education continued in a small office among Manhattan’s wholesalers and rug dealers. Matt Philips handed him a fifteen-page specification for an ad server. Six weeks later, he demonstrated a prototype that did what the document asked. Mike Walrath, the company’s founder, disliked it. Following instructions had produced the wrong answer.

O’Kelley recalls asking Walrath to explain what he actually wanted. Walrath was manually studying reports and calling publishers, trying to put the most productive advertisement in the right place. The software needed to reproduce that judgment. O’Kelley began thinking about ranking ads by expected value rather than a predetermined order. An auction could replace the queue.

“I was going to build a software version of Mike’s brain.”

Brian O’Kelley, recalling Right Media

The conversation also changed the team. In O’Kelley’s account, Philips subsequently proposed that he become CTO, accepting a change in their reporting relationship. The detail matters because the origin story contains more than a clever algorithm. Somebody had to admit that the specification needed changing, and somebody else had to make room for the person who could change it.

A conference pass, four networks, and breakfast diplomacy

The next opening came while O’Kelley was walking the exhibit floor at Ad:Tech in New York in 2004. Right Media could not afford a booth. A conversation with product manager Elad Efraim introduced a possibility O’Kelley says he had barely considered: licensing the technology to other ad networks. A company that sold advertising could also sell the machinery for selling advertising.

Walrath had an understandable objection. Why let competitors use the algorithm on which the business depended? O’Kelley argued that a European customer could bring both software income and opportunities to trade traffic. The proposed deal required rebuilding the platform for multiple networks. The commercial idea arrived before the architecture was ready to receive it.

His recollection of the 2005 rollout is less polished than a product launch presentation. There were reporting errors, failing servers and unhappy clients. He brought pastries from City Bakery to meetings with the Oridian team, listened to the complaints, then carried the work back to his engineers on Post-it notes. Croissants were doing a modest amount of account management.

Four networks went live on Yield Manager on April 1. After the initial trouble settled, he saw advertisers getting better results, publishers earning more and the network making more money. Adding networks had widened the pool of advertisements available to match with an audience. What first looked like another reporting mistake became evidence that the exchange could work.

The exit, and the detour through things you can touch

Yahoo acquired Right Media in 2007. That year, O’Kelley co-founded AppNexus, spending the next eleven years as its CEO. Where the earlier business had exposed the possibilities of an exchange, AppNexus made independent advertising infrastructure his principal occupation. The engineer had become responsible for a company as well as its machines.

AT&T acquired AppNexus in August 2018 in a deal widely reported at about $1.6 billion. O’Kelley stayed as a strategic adviser into early 2019. Acquisition prices make convenient punctuation in a founder’s biography, but they obscure the more interesting transition here. Having spent years making digital transactions move faster, he went to work on transactions involving physical materials.

In 2019 he co-founded Waybridge, a supply-chain technology company for commodities, and served as CEO until December 2021. The move took him into a business where logistics has weight, inventory takes up space and a shipment cannot be refreshed into existence. Supply chains became a practical subject rather than a metaphor for the route an advertisement takes.

A supply-chain course at MIT supplied another provocation: a banana’s carbon footprint changes with its destination. O’Kelley became interested in what happens when the cost of an object includes its journey. Applied to advertising, the question becomes awkward. An ad looks weightless on a screen. The computers carrying out the transaction are plugged into something.

The invisible delivery bill

Scope3 launched publicly in January 2022 as a public benefit corporation. Its founding group brought together engineers and advertising veterans, including Anne Coghlan. The ambition was to measure emissions along the advertising supply chain and give buyers information they could use when allocating money. A familiar market could be examined through a different unit of account.

The name came from the accounting category for emissions across a company’s value chain. Scope3’s methodology followed the path of an ad, including where it was served, rather than stopping at a company’s own operations. This made the digital supply chain visible as a series of energy-consuming activities. A campaign report could contain an environmental dimension alongside its commercial measures.

For O’Kelley, the appeal was partly economic. He argued that advertisers could improve what they bought while eliminating waste. In 2022 he described a future built around fewer, better advertisements. It was a conspicuous change of emphasis for someone associated with expanding the market’s capacity to transact. Getting more machinery involved in an ad had ceased to be an adequate definition of progress.

In October 2023, Scope3 raised a $20 million Series B led by GV, with Room40 Ventures and Venrock also participating. The funding backed its collaborative sustainability platform and emissions-reduction tools. O’Kelley’s engineering knowledge was useful in an unusual way: he understood the system well enough to investigate the bill it had been leaving out.

Then the machinery changed again

By March 2025, Scope3 was introducing an Agentic Media Platform and Brand Standards, moving into AI-based advertising decisions. An agent, as O’Kelley explained it that summer, is software that can act with some degree of autonomy on somebody’s behalf. The proposition extended beyond measuring a transaction to changing how a decision gets made.

In a conversation with Index Exchange, he described moving decision-making closer to the supply of advertising. The reasoning was practical: keep useful information near the transaction rather than stripping away context before making the choice. He spoke about building on systems already in use, then testing whether additional intelligence could improve their results. The first hurdle was simply getting the pipes to work.

The September 2026 Apostra launch made the new direction explicit. Coghlan described a gateway connecting brands, media owners, creators and their agents. Instead of expecting every advertising opportunity to fit the same transaction, the company wanted agents to work across different media and commercial arrangements. A billboard and a social placement could keep their differences while becoming easier to discover and buy.

O’Kelley’s accompanying explanation divided the challenge into connection, orchestration and governance. Getting agents to communicate was one job; helping them cooperate with people was another. Making sure their actions stayed within human intentions was a third. His own list included the wonderfully unceremonious concern that agents might screw everything up. Infrastructure has a way of making that possibility concrete.

Letting other people into the workshop

A common language sits underneath this approach: the Ad Context Protocol, or AdCP. The open specification gives advertising agents a shared way to communicate. O’Kelley serves as its lead architect. The distinction between a company’s product and a shared protocol matters to his argument. Competitors need a reason to use the language even when they choose somebody else’s services.

He has also been arguing publicly about when to open the workshop door. In April 2026, responding to Andrew Kraft’s concern that a premature standard could constrain invention, he defended developing AdCP in public. Let implementers find faults while the design can still change. He also acknowledged the danger of a project depending too heavily on its original architect.

“Some of the abstractions in 3.0 will turn out to be wrong.”

Brian O’Kelley, on building AdCP in public

That admission gives his latest chapter an interesting connection to the fifteen-page specification at Right Media. A coherent design can still misunderstand the job. The remedy he proposes now makes criticism part of the process, rather than something encountered after a finished product has been unveiled. The willingness to revise has become an explicit piece of the architecture.

In October 2026, O’Kelley and Coghlan were writing about buying platforms that could span advertising formats and coordinate work from planning through reconciliation. It remains a proposal for where the industry should go. His life outside that argument is in Brooklyn, where his public biography places him with his wife, two daughters and two cats. Advertising may be increasingly automated; the household has several constituencies of its own.

The career’s recurring scene is a builder discovering that the job has changed. An ad server becomes an auction. A supply chain becomes an emissions map. A carbon-measurement company becomes a gateway for agents. The unfinished business is deciding which work deserves to survive the next revision. O’Kelley has spent much of his career increasing what advertising software can do. He is still negotiating what it ought to do.