Shared software for independent newsrooms

What I remember from 2004 is a conference room at Cox Newspapers in Atlanta with a whiteboard full of boxes and arrows, three producers arguing about where a shared photo service should live, and a systems engineer quietly working out how we would keep eighteen newspaper sites alive through a hurricane season without falling over. We were building a platform for a company of newspapers that had, until recently, each built their own everything. In March 2005, two of our projects received NAA Best Practices awards. The awards were nice. Keeping those newsrooms working was the achievement.

Shared infrastructure, local control

COXnet was the digital group inside Cox Newspapers. The Atlanta Journal-Constitution, the Palm Beach Post, the Austin American-Statesman, the Dayton Daily News, and fourteen other daily and weekly papers each had their own site, their own voice, and their own classifieds relationships. The problem we were solving was not how to make them look the same. It was how to give them shared plumbing so that each paper could spend its scarce web staff on local journalism and local advertiser relationships instead of rebuilding CMS features from scratch eighteen times.

The architecture we settled on was a shared content platform with local branding and local editorial control. Story publishing, photo handling, user registration, advertising insertion, classifieds integration, search, and email newsletters ran on a shared codebase. Look and feel, section hierarchy, and editorial judgment stayed with each paper. In 2004, a lot of chains were still running a site-per-paper model and quietly going broke maintaining it.

The photo service nobody wanted to build twice

Our shared photo and image service saved work across the newspapers. Photographers at Cox papers were shooting digital, wires were dumping hundreds of images a day, and each paper's web team was resizing, captioning, and publishing images through a workflow that looked a lot like a scream. Our team shipped a centralized image service that handled ingest from the wire, ingest from staff photographers, automatic resizing for the templates each paper used, caption persistence, and rights tracking. Once a photo was in the system, any Cox paper could pull it, any Cox site could render it at the right dimensions for the right slot, and we stopped paying for eighteen copies of the same resize logic.

The piece of this work I am still proud of is the resize pipeline. We treated image derivatives as a cache of renderings, not as separate assets. When a paper redesigned a section and needed a new thumbnail size, we did not ask photographers to re-upload. The pipeline just generated the new size from the source on demand and cached it. It saved people a great deal of work at the time.

Classifieds that did not break the paper

The shared classifieds work lived closer to the business. Classifieds were still real money for newspapers in 2004, and our papers had longstanding relationships with Cars.com and real estate partners through Classified Ventures, and with CareerBuilder separately. The problem was integration. Each paper had been wiring up these feeds differently, which meant every partnership change rippled through eighteen codebases and eighteen launch dates.

Our team built a shared classifieds integration layer. Listings came in through a common ingest, local papers could apply their own presentation and their own local listings on top, and partner changes landed in one place. The less glamorous part, and the part that actually mattered, was the monitoring. We instrumented the feeds so that when a partner's nightly dump went stale, a page stopped updating, or a listing count fell off a cliff, somebody was paged before readers noticed. Most of the art in this kind of infrastructure is not in the normal path. It is in the detection path when something goes wrong.

The two awards

Our team at COXnet built a way to upgrade software while the newspapers kept publishing. The Newspaper Association of America's Presstime magazine recognized our live software upgrades in the pre-press category of its March 2005 Best Practices awards. Its account of the work describes a system that lets Cox's 22 newspapers continue publishing while upgraded software replaces the running version. That is a more useful accomplishment than keeping an unchanged website on the screen.

Julie Leddon, our manager of product development, explained the benefit in the same report: enhancements and fixes could reach the newspapers without requiring all 22 to agree on a convenient shutdown. The engineers who made that possible were solving a problem for the people using the software, as well as for those maintaining it.

The COXnet Suite of Print+Web Applications received the new-media award. Print+Web let writers and editors enter and edit content once for both print and web publication. It also made information searchable and linked: sports scores, statistics and standings could update online, alongside box scores updated on game nights, while calendar listings carried web links. A local restaurant database was still coming soon.

What continues during the change?

An editor with a correction to publish should not have to wait for a software upgrade to finish. Keeping the existing story available to readers is useful, but it does not solve the editor's problem. The publishing system has to accept the correction and get it onto the page.

A maintenance window is an agreement to stop some work. If a change requires one, finding a tolerable time is necessary. But that negotiation starts with an assumption: people will have to stop using some part of the system. I would rather question that requirement early, when there is still an opportunity to change the design.

This is especially worth doing for a shared publishing system. A quiet period at one newspaper need not be quiet at another. Even if everyone can agree on a time, a release schedule is no reason to expect the news to cooperate. The attraction of live upgrades is that each newsroom can continue making its own decisions about when to publish.

To evaluate such a design, I would start with an ordinary piece of editorial work. Suppose an editor opens a story before an upgrade starts, then saves a correction after the change. Which software handles the save? Can it read the work already in progress? Does the corrected story reach the website? Those are questions about the whole path from the editor's action to the reader's page. A successful request for an existing page answers none of them.

There can be changes for which an interruption remains the responsible choice. If new software changes the meaning or arrangement of stored information, allowing the old and new versions to use it together may require more work than replacing the programs themselves. Calling the release live does not settle that problem. The engineering decision is whether the necessary work can continue safely, and, where it cannot, how narrowly the interruption can be confined.

Recovery includes the work done along the way

Continued publication makes recovery a more demanding question too. Returning to the previous software version, often called a rollback, is only part of it. The newspaper may have published new stories and corrected old ones since the upgrade began. That work needs to remain usable if the software changes back.

Consider a new version that stores a story in a form the previous version cannot read. Putting the old program back would not, by itself, restore a working publishing system. Restoring an earlier copy of the data might make that program happy while discarding the newsroom's subsequent edits. Neither is the result an editor means by getting back to work.

I would want a recovery plan to account for that interval explicitly. Start with a story before the change, change it during the upgrade, and then exercise the route back. Check the resulting story, not just whether the program starts. The test should tell the newsroom what it can rely on if a release has to be reversed. It should also expose a change that cannot safely be reversed, while there is still time to choose a different release plan.

The ability to change software is part of the software's job. If that ability is designed and tested alongside the publishing features, the team has fewer reasons to ask editors to stop using those features whenever an improvement is ready.

What I value in the COXnet work is the room it gives the newspapers to keep working. A useful correction can reach a reader while the engineers are still making the system better.

September 2026 reflection

I later came to believe, and still believe, that most things we call best practices are really good practices in context. Cox's context in 2004 was a chain of mid-sized to large metropolitan newspapers, each with a strong local identity, each answerable to a local publisher, operating under a parent company that was willing to invest in shared digital infrastructure. The same architecture would not have worked at a single-paper company, and it would not have worked at a chain where the papers did not trust each other. It worked for us because the business shape fit the technical shape.

The projects that won those awards were good practices that fit Cox's shape. The deeper win was not the architecture. It was that the people building it were willing to share credit across eighteen sites, argue in public about trade-offs in that Atlanta conference room, and keep the newspapers functioning while the plumbing got replaced under them. I got to lead a slice of that work. The people I want named in any honest retelling are the producers, developers, systems engineers, designers, and editors who did the day-in-day-out work of making one platform serve many newsrooms. The plaques sit in boxes somewhere. The habits that team built are what I took with me.

A few years later, I was surprised by how much of the payoff was cultural rather than technical. By building one platform together, the papers started to share more than code. Section producers from different cities started comparing notes on what was working. Engineers loaned each other capacity during big news events. When a hurricane knocked down infrastructure near one paper, the others could and did carry load. None of that was in the architecture document. All of it was what the architecture made possible. The team we built, and the platform that team built, stayed useful long after the press releases had yellowed.