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.
Danger as marketing
Last week Cal Newport called superintelligence a fairy tale. This week Paul Worthington explains why the labs keep telling it. Framing, not targeting, is Silicon Valley’s real superpower. Google didn’t sell search ads, it “organized the world’s information.” Uber wasn’t a taxi company, it was “ridesharing.” Now listen to the AI vocabulary. Models don’t return wrong answers, they hallucinate. Agents go rogue. Systems escape. Nobody talks about a spreadsheet that way. Worthington’s read is that the doom frame solves a business problem. Treated as plain software, frontier models look like a commodity: open-weight models trail by months, distillation pushes capabilities downmarket, and prices keep falling. It’s hard to see how the next $100 billion buys $100 billion of durable advantage. But if the technology is existentially dangerous, restricted access becomes the “responsible” position, and the companies best positioned to comply are the ones already spending the $100 billion. “The harder it becomes to sustain scarcity through the technology itself, the more economically valuable a frame becomes in which unrestricted access to that technology is perceived to be dangerous.”
— Off Kilter, 12m, #ai, #strategy, #marketing
Rationing tokens
A year ago every CIO wanted AI adoption at any price. Now they’re handing out token limits, but no one knows what the limits should actually be. TIAA’s Sastry Durvasula went job function by job function, guessed a ceiling for each, and built a review process for anyone who hits it. Some requests get approved. Many don’t: “on research teams some of the analysts are just burning their tokens.” Carvana refused to set hard caps at all. Alex Devkar publishes monthly “expectations” in dollars, sends weekly progress reports, and treats a spike as a reason for a conversation, because it might be a runaway process or it might be “a really rich vein and we should lean in here.” Principal Financial set one hard limit for everyone, then had to raise it after people started rationing and holding back. The bottom line: you can’t tell waste from breakthrough by reading the meter, someone has to look at the work.
— The Wall Street Journal, 5m, #ai, #management, #strategy
Twenty-eight times the spend
DX’s latest report found that quarterly AI spend at technology companies went up 28X in twelve months. But over the same year, the share of engineering effort going to new features didn’t move. Chantal Kapani calls this the “AI efficiency paradox.” If an engineer produces a pull request twice as fast and it then sits waiting for review, the organization hasn’t gotten faster. It has just moved the bottleneck. Review queues, CI delays, security sign-offs, and fuzzy requirements now absorb the hours the coding assistant saved. Atlassian’s numbers say the same thing from the top: only 6% of executives can confidently point to organization-wide AI ROI, even though 90% of organizations are now paying for models. The freed capacity is real, but it’s being spent on waiting.
— LeadDev, 5m, #ai, #engineering, #productivity
Five lines of code
Every developer can call a frontier model in five lines of code, and that’s the problem. The API hides the model. Sebastian Raschka, who wrote the book on building an LLM from scratch, makes the case for developers to look at the inner workings anyway. Not so you can train one, but so you can be “more critical of the AI’s outputs and approaches.” His example is watermarking. Anthropic says it can stamp Claude’s text. Does that hold up? If you know how a model picks each token, including the dice roll at the end, you can answer that yourself. If you don’t, you’re taking a vendor’s word for it. Same with reasoning models. Know that, and you know when a long chain of “thought” is real work and when it’s a show. Agents write more of the code every quarter, so review is the job now, and you can’t review what you don’t understand.
— Tech World With Milan, 9m, #engineering, #ai, #leadership
Politics is the job
Gregor Ojstersek’s first lesson in company politics came in a meeting where engineering blamed product’s requirements and product blamed sales’ promises. His second came when he tried to fix a process outside his lane and got told to keep his nose out of it. The best technical answer, he concluded, rarely wins on its own. Now his playbook relies on deliberate relationship work. Learn how your peers like to work before you propose a process for them. Draw an “influence diagram”, because the org chart lies about who actually decides. Invest in relationships early, because when a peer weighs your proposal, the questions in their head are “how risky is this for me?” and “what do I gain or lose?” Build public credibility, which he characterizes as armor. And don’t play the dysfunctional game with public blame, 11 p.m. urgent emails, performative availability. Treat politics as part of the role rather than a distraction from it.
— Engineering Leadership, 8m, #leadership, #management, #career
Echo of the week
Echoes are AI agents in Steady that automatically gather and deliver work context to teams on a schedule – answering recurring questions about progress, capacity, and coordination so you stop burning hours assembling the same information manually.
Decisions waiting on you is for the week you realize the bottleneck LeadDev describes has your name on it. Every decision routed through you is a small tax on everyone waiting for it. Each one feels like leadership in the moment, but in aggregate they turn you into a queue, and the work stalls at exactly the rate you can clear it.
Every Wednesday, this Echo surfaces the approvals, decisions, and blockers that appear to be stuck on you specifically, notes who’s affected, and asks whether the person closest to the work could make the call instead. Most weeks, the honest answer for several of them is yes.
The human-agent teamwork OS
Teams rely on two coordination loops to function: a big-picture loop connecting plans to progress, and a ground-level loop keeping teammates in sync.
Running those loops was already a marathon of meetings, chat threads, dashboards, and manual toil. Pile on flatter orgs, exponential output, and AI agents shipping 24/7 — the old way can’t keep up.
Steady removes the coordination bottleneck by running both loops for you. Working in the background, Steady distills updates and activity into targeted context for everyone on the team — human and agent alike. Full visibility, tight alignment, zero overhead.
The outcome: high-performing teams that deliver the right work, not just more of it.
Learn more at runsteady.com.