Not Just Faster Horses

Jenni Ryall founded Other Labs, a media capability and AI advisory working with newsrooms, funders and platforms across Australia, New Zealand and the Pacific. Her two decades in media include Mashable in the social media era, TEGNA in US local news and Meta during the News Media Bargaining Code. She now runs a Google AI program for publishers, currently in its fifth cohort.

This is the story of why media finds AI hard, and what the newsrooms getting results do differently. Many legacy publishers run on decades-old systems and never finished going digital, and the industry has been burned by technology before, from the pivot to video to the social media era. Jenni explains why small newsrooms often move faster and walks through Gazette, an Australian startup that built a tool to turn council videos, documents and emails into stories, with an editor reviewing each one. Over 12 weeks, its story output went up about five times. She also covers 6AM City, which uses AI newsletters to decide where to put human teams.

Then there’s the program itself. Every project now starts from a time spent and accuracy baseline, often set by timing someone doing the task by hand. Her latest program runs for 18 months with technical support, because some people want to direct the work and leave the building to others. Each newsroom picks one idea with metrics attached and checks in weekly with a mentor, which matters when AI makes it easy to build the wrong things faster.

We also cover the people side: using staff who are deep into AI as early adopters, the efficiency trap of working at 100 times your old pace, and why leaders should pull those people into strategy work. Plus the policy question most organisations haven’t answered: who owns a company AI account, and everything in it, when someone leaves. To finish, Jenni talks through what she’d build if she started a newsroom from scratch, the AI UX slop that makes a business look synthetic, and why she’s still a tech optimist.

Nick Torres is CTO at Firecrown Media, a challenger media company with more than 50 websites across 20 brands, in verticals from hobbies and travel to freight and logistics. The engineering team behind all of it has four people, including Nick.

This is the story of how Firecrown got the whole company building with AI. Every employee got a Claude seat, and the first AI town hall went quiet. The shift came from one question: which parts of your job do you hate? The HR lead now supports about 350 staff across three businesses with apps she built herself in Claude and Replit. A sports editor's four-hour weekly stats job takes about an hour. The CRO took over the homegrown CRM and prefers it to the vendors he's been demoed.

Engineering changed too. The team migrated 28 websites in three days. Every project lives in GitHub, so each migration becomes an agent the team reuses on the next site. Nick now sees his developers as product managers, with their Claude instances as their teams. A WordPress plugin for Firecrown's entitlement system spent two years on the wish list and is now in production.

Then the numbers. Firecrown spends close to six figures a year on Claude and Replit across the company. Nick valued the apps built so far at about a million dollars. One expense system alone saves about $100,000 a month in budget leakage.

We also get into the two questions Firecrown asks before building anything, why email delivery got vetoed, and taking over the Times Free Press with a 90-day handover while running 45 days ahead of schedule. 

Most media companies have given their engineers AI tools. The engineers are finishing work faster. And delivery hasn't moved. Stéphane Boisvert knows engineers who are idle for half the day, waiting on everyone else.

Stéphane has spent 15 years inside enterprise WordPress, including time at WordPress VIP and as an agency CTO. He now runs Principal WP, where he works with media and engineering teams on deploying AI across the whole delivery process, not just the coding step.

His argument is that most teams have built a faster horse. They've sped up one person at one step, and left every gate, handoff and unanswered Slack message exactly where it was. PR review times are up. Bugs are up. The bottleneck just moved.

In this conversation, he gets specific about what the alternative looks like:

We also get into the IT manager who wants two weeks to rebuild HubSpot, where WordPress sits as vibe-coded sites take the small end of the market, AI finding security bugs in decades-old open-source code, where the junior engineer of 2030 comes from, and why open-source projects will need to ship context and skills for agents, not just code.

Alison McHenry is CTO at Daily Kos, a community-funded progressive media site with 20 years of content and 88 million comments. When she joined in 2024 the company had over 100 people. It now has 26, with one in-house engineer alongside Alison and an outside delivery partner.

This is the story of what that team did with those numbers. Daily Kos was running a custom CMS split between Perl and Ruby on Rails, with no tests, no analytics and no observability. Nobody had a mental model of the whole thing, the search engine failed half the time, and the AWS bill was the biggest bill in the company. Alison walks through the nine-month migration to WordPress, why she reckons it would have been a two-year job with a team of 25 a few years ago, and how AI was used to reverse-engineer a codebase nobody fully understood: consolidating 67 user roles down to five, working out which pages were even alive, and deciding what didn't need to come across at all.

Then the launch. A two-week standalone beta with 500 users found the WP admin left open to anyone with the URL, duplicate email records, and a commenting system that didn't thread the way the community needed. The fix was a CSS layer over the vendor's output, because waiting on the vendor wasn't an option.

We also get into the engineering process underneath it: AI writing code and humans doing the review, keeping pull requests small enough that approval isn't a rubber stamp, CodeRabbit as the first pass, and why Alison still writes every product brief herself before handing it to Claude. Plus the build versus buy calculus for a small publisher, the SaaS fallacy she's watched media companies fall for, and what it means to ship a save-articles feature in two and a half weeks when it would once have been a six-month project.

Recorded at WordCamp US in Phoenix.

Seth Rubenstein runs engineering at Pew Research Center, one of the most cited research organisations in the US. His team is six engineers. Their day-to-day is managing agents, and they track what that actually costs: around $4,000 to $5,000 a month in tokens, a number that went down when they started testing which models earn their keep.

This is the conversation about what that looks like in practice. Seth pushes back on the idea that AI is replacing headcount. At Pew it's the opposite: agents covering roles a six-person team never had, like accessibility review, for less than a fraction of a salary. He walks through the WordPress VIP partnership that compressed a year's roadmap into a few months, why Pew wants ChatGPT indexing its content when most publishers are fighting it, and the moment they realised LLMs were inventing plausible-looking Pew URLs that 404.

We also get into the workflow change underneath it all: hallway conversations and quick prototypes replaced by tickets assigned to agents that come back with screenshots proving what they did. And the question the whole industry is about to hit. When you can build anything, what should you build?

Recorded during WordCamp US in Phoenix.

What's covered

Why "AI replaces headcount" isn't playing out at small organisations. Real token budgets and how they benchmarked models against cost. The Word to WordPress project, collapsing print-first production workflows. Getting AI-ready in 30 days with WordPress VIP. Accuracy over traffic, and hallucinated URLs. Why juniors still need fundamentals. Whether you can vibe code a CMS in an afternoon, and what happens after you do.