Steady The Steady Beat newsletter

The hard part The Steady Beat, Issue #122: what Revolut learned when tokens got cheap, slow teams vs. slow organizations, trust as the bottleneck, the decisions that are the job, and why floppy disks aren't the FAA's real problem. You're reading The Steady Beat, a weekly pulse of must-reads for anyone orchestrating teams, people, and agents across the modern digital workplace — whether you're managing sprints, driving roadmaps, leading departments, or just making sure the right work gets done. Curated by the team at Steady. Unmetered Revolut cut its AI coding bill by about 90 percent by running an open-weight model on its own GPUs, but the big win came after engineers stopped rationing. David Abitbol, who runs ops and infrastructure there, says the company first rolled out AI coding tools to about 2,000 engineers with strict budgets, because some were spending $1,000 to $10,000 a month with little to show for it. Self-hosting GLM, a Chinese open-weight model, was far cheaper than Claude Code or Codex. Once tokens were effectively unlimited, engineers pointed AI at debugging and Jira and Confluence tickets, not just new code. That's where the biggest gains came from. "You no longer have to save your tokens for the end of the quarter or the most difficult task." — The Register, 4m, #ai, #engineering, #costs Faster chaos When a team looks slow, the problem is often the organization around it. In 2017, Jekaterina Jegoscenko took over a development team everyone thought was underperforming. The engineers were good, but the system around them wasn't. Requests arrived by email, spreadsheet, and hallway conversation, nobody owned priorities, and Scrum meetings happened without anyone remembering why. So she fixed the system. She gave requests one way in, took over prioritization so developers stopped negotiating with five managers, and showed stakeholders working software instead of status reports. Trust came back, and so did the team's motivation. Nearly a decade later, she sees the same problems in teams that use AI: unclear ownership and conflicting priorities. The difference is speed. Developers once spent days building a request nobody prioritized. An agent does it in minutes. "If we automate execution without improving decision-making, we risk producing the wrong things at a much higher speed," she writes. — Jekaterina Jegoscenko, 12m, #management, #process, #ai Tweak paradox Teams now write code faster than they build trust, and Kent Beck says that gap is where projects fail. Trust comes from slow, mostly invisible work: testing, review, pairing, and shared understanding. AI sped up the visible part, the code, and teams now give up those practices to move faster alone. Beck's main point is what he calls the "tweak paradox." The bigger a change, the less you can trust it, and trust drops much faster than size grows. AI makes it easy to produce big changes. Trust demands small, verified steps. He also warns against the pull toward everyone working alone with their own agent: "it is a huge mistake." His answer is to slow down in the right places, because that's how you go faster. — Kent Beck, 3m, #engineering, #ai, #teams The decisions are the job When agents write the code, an engineering leader's value is in the decisions around it. LeadDev's Lucy Adams asked three engineering leaders which skills matter most now. "AI dramatically reduces the cost of producing an answer, but it hasn't reduced the cost of making the wrong decisions in production," says Sreenivasa Reddy of BNSF Railway. Checking what AI produces is real work that rarely shows up in metrics. Juniors are losing the grunt work they used to learn from, so Stanford's Matt Vassar says to let them write the tests an agent has to pass. Former Yahoo manager Tiffany Jachja fights the urge to jump in and fix things by asking what she's taking from someone else. When nobody knows what the team looks like in five years, Reddy says to share what you know, what you