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.
Kodak moment
DHH built his reputation on software craft, so his Rails World keynote hit the audience (and the Rails community) hard: “At 37signals, a couple of weeks ago, we made the decision that it clearly means we’re done writing code by hand.” Hand-written code is now the exception there, treated like a red flag. He compares the moment to what photography did to painting, and he dates it to November 2025, when Opus 4.5 shipped: “the ‘Kodak Brownie’ of our era.” He says he’s retired as a professional programmer, and that “writing code by hand is no longer an economically productive enterprise for the vast majority of programmers.” Gergely Orosz compares the proclamation with less cheerful reports. One big tech engineer describes 12- to 13-hour days of shipping AI-generated code with little quality control: “Nobody is thinking anymore. Everything is done by LLMs.” Uber Eats shipped a new add-ons selector with three obvious bugs, including an offer to “choose up to 999” toppings. Orosz expected agents to reduce engineers’ work, but he says it’s growing instead: engineers have to learn how LLMs behave, build systems that validate what agents produce, and build new kinds of products.
— The Pragmatic Engineer, 15m, #engineering, #ai, #leadership
Notation up
Sam Ruby agrees with DHH that agents will write the code. His question is what language they should write it in. To find out, he compiled Campfire, the 37signals chat app, unmodified, from Ruby on Rails to C. The result is a single 202,000-line file that needs no Ruby runtime and runs at 7 to 8 times the throughput in 92% less memory. It’s also waaaay too big for an agent to work in. The Rails app is about 60,000 tokens and fits in an agent’s context. The C version is about 4 million. From that comparison, Sam lists what source-of-truth code needs: it fits in context, says each thing once, keeps the intent, produces the same result every time, and looks like something the model has seen before. Rails passes all five. So agents edit Rails, and a compiler turns it into whatever runs fastest. In Sam’s words, “Rails is the notation; what runs it at the bottom is a separate question.”
— Sam Ruby, 9m, #engineering, #ai, #systems
Invisible work
Two years ago, programmers asked “who am I if the genie [AI] codes?” Now mathematicians are asking the same thing about proofs, and Kent Beck has a way for them to think about it. Every field has visible work and invisible work. In software, the visible part is features. In math, it’s proofs. The invisible part is, in Beck’s words, “understanding, education, simplification, enabling abstractions.” AI is great at the visible stuff. But “without the invisible work, visible progress eventually slows to a crawl, genie or no genie.” Beck’s own answer as a programmer: “I’m here to keep the genie on course.” The genie helps him learn faster, his intuition kills dead ends sooner, and “my reach has just extended.” He works in stretches, visible progress first, then a break to make invisible progress. “Yes, my work has changed, but my strategic decisions are more valuable than ever because they come more frequently.” He hopes the same framing lets mathematicians use the new tools “without losing their reason for being mathematicians.”
— Kent Beck, 4m, #ai, #engineering, #systems
Skill issue
Vercel’s skills.sh registry launched in January and passed one million agent skills in seven months, with nearly 280 million installs. GitHub took 27 months to reach a million repositories when it launched. npm took more than nine years. A skill is a set of reusable instructions an agent follows for a particular job. Engineers write most of them: over half the listings cover software engineering, data, infrastructure, security, or agent workflows. But people in every kind of job install them. Compared with the average skill, a business operations skill gets 74% more installs and a writing skill gets 50% more. A software engineering skill gets 30% fewer. A few skills get almost all the use. The top 1.2% account for 94% of installs, and nearly half have been installed once. Vercel predicts that public skills become something every company has, and the advantage moves to private skills that capture a company’s own judgment, like who qualifies for a refund or what’s ready to ship.
— Vercel, 5m, #ai, #productivity, #strategy
Who decides
Researchers at MIT’s Center for Information Systems Research say “the same AI capability can be safe in one context and risky in another,” so companies should sort decisions before handing them to agents. They sort on two questions: how clear is the right answer, and how bad is a wrong one? That gives four kinds of decisions, and the telecom One New Zealand has an example of each. Routine decisions are clear and low-stakes, so agents do the work while people stay close. One NZ’s agent writes the database queries that pick a marketing campaign’s audience, cutting that work by 60%. Consequential decisions are clear but costly to get wrong, so people monitor and handle the exceptions. At One NZ, every pricing decision needs a human, because accurate pricing is a “nonnegotiable constraint.” Exploratory decisions are open to interpretation but cheap to get wrong, so agents can experiment under human oversight, like drafting marketing content. And strategic decisions are uncertain and high-stakes, so people lead and AI supports. One NZ’s network planning trades off reliability, customer experience, cost, and infrastructure, so humans make that call.
— MIT Sloan, 6m, #ai, #leadership, #governance
Peer sport
At most companies, feedback flows down the org chart and shows up at review time. Thoughtworks’ Anuja Karnik and Sumeet Gayathri Moghe want it to move sideways, all the time. They treat feedback as a peer sport and as telemetry, a steady signal about team health that warns you before things break. To give it, use Situation-Behavior-Impact: describe what you saw and what it caused, and leave out what you think it says about the person. Ask why before you judge, ask permission before you start, and deliver criticism privately and live, because text strips out warmth. To receive it, there are only two responses, per John Reid-Dodick: “Thank you,” and “Thank you, tell me more.” Then close the loop and tell the person what you did with it. But watch out for antipatterns, like public 360s and “speedback” sessions that trade honesty for theater. And tying peer feedback to performance reviews puts salary on the line, which kills candor.
— Martin Fowler, 20m, #leadership, #management, #coordination
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.
Team retro prep. Retros run on memory, and memory favors whatever went wrong last Tuesday. Once a month, on a Wednesday at 2:00 PM, this Echo reads the last 30 days of check-ins and goal updates and gives you the raw material instead: every goal the team is working on, the month’s progress in five themes with a paragraph each, and the top five trends in blockers and challenges. You walk into the retro arguing about patterns, not anecdotes.
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.