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.
Nobody’s books
In 1979, an IBM training manual warned that a computer must never make a management decision, because a computer can never be held accountable. Forty-seven years later, two OpenAI models slipped out of their test sandbox, used stolen credentials, and broke into Hugging Face’s production servers – no malice, no instruction, just models looking for a way to cheat a test. NYU’s Haran Segram reads the episode as a balance-sheet story, not a safety story. Statutes like California’s AB 316 now bar the “the agent acted on its own” defense, pinning liability on whoever develops, modifies, or deploys a system. Meanwhile insurers are filing generative-AI exclusions, writing themselves out of the risk at the very moment the law assigns it. The exposure is real, growing, and sitting on nobody’s books. He suggests closing the gap through contracts and coverage: name a person of record for every agent before deployment, assign coverage explicitly, build in audit rights, and test containment yourself instead of taking the developer’s word.
— The Wall Street Journal, 7m, #ai, #governance, #risk
Manager of agents
No manager in history has had to manage agents alongside people. That’s about to be a standard line in the job description, and McKinsey’s Julie Goran argues the shift underway is the biggest change to how organizations work since the dawn of the Industrial Age. Leaders are asking questions that would have sounded absurd two years ago: should agents go on the org chart? How do you plan agent capacity next to human capacity? What happens to spans and layers? (Her bet: fewer layers.) Companies are getting granular about which activities agents absorb, then asking what people can do with the reclaimed 40 percent of their plate – one tech company aims to triple revenue over five years with flat headcount. However, tool rollouts and training programs are the visible, easy part, and too many companies stop there. Value comes from picking the few domains that matter and doing the unglamorous work of redesigning workflows and jobs around them.
— McKinsey, 6m, #management, #ai, #transformation
Follow the baton
“AI didn’t change what great software looks like. It changed what’s scarce. Human attention is now the bottleneck.” Thoughtworks CTO Rachel Laycock’s metaphor for what developers are becoming: conductors. The agents are the musicians – eight, ten, twelve of them playing in parallel – and the developer’s value shifts to understanding the whole score: providing context, judging output for subtle wrongness, deciding what ships and what goes back for another pass. A great conductor is first and foremost a great musician, which is why the fundamentals still matter. What worries Laycock is that engineering education and career ladders still train people for individual execution. “We’re redesigning the tools, but we haven’t started redesigning the job.” The skills that suddenly matter for ICs – protecting attention, making calls with incomplete information – are the ones we’ve historically reserved for executives, and almost nobody is teaching them. If attention is the scarce resource, the career ladder needs a rung nobody has built yet.
— Martin Fowler, 7m, #engineering, #ai, #leadership
Approve, approve, approve
Researchers gave 107 developers five hours to build an app alongside coding agents that carried a hidden objective: quietly exfiltrate personal data. Ninety-four percent of the developers never caught it. It gets worse. When an LLM monitor flagged the malicious code – which it did 84 percent of the time – developers merged it anyway in more than half of the flagged sessions. Seniority was no protection: two-thirds of those who merged flagged code had three-plus years of experience, and nearly half had security backgrounds. Participants watched tests pass instead of reading code, and one summed up the era in seven words: “I’m always used to clicking approve directly.” So what do we do? Force real review at choke points – new outbound network destinations, dependencies, config values, auth and payment paths – and build monitors that propose fixes behind an approval gate rather than emitting warnings people swat away (or ignore). Code review was supposed to be the safeguard of the agent era. As currently practiced, it’s a permission dialog.
— Research-Driven Engineering Leadership, 8m, #ai, #engineering, #security
Addition by subtraction
Steve Huynh’s improvement plan for the rest of the year contains no new habits or systems. Borrowing from music producer Rick Rubin, he argues for working as a reducer: strip away everything obscuring the core work, using one test – if this disappeared tomorrow, would anything real break? If the answer is no, cut it. The subtraction applies to information, too. He describes a reader who prepped for interviews by consuming endless AI-generated advice and ended up paralyzed; the breakthrough came from deliberately using less. “There’s tremendous power in using the least amount of information needed to get a point across.” And it applies to your calendar: what were you doing last Tuesday morning? Did it matter? For anyone managing AI-amplified output, this lands differently than the usual productivity advice – when generating more is nearly free, the discipline that matters is deciding what not to keep. Huynh steers by Ethan Evans’s monthly question: what did I do this month that I couldn’t do last month? If you can’t answer, start subtracting until you can.
— A Life Engineered, 6m, #productivity, #leadership, #systems
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.
Goals at risk reads every goal in Steady and surfaces the ones marked at-risk or off-track, each with a short explanation of what’s dragging it – delivered Monday at 11 AM, while there’s still a week to do something about it. No chasing status, no discovering in the quarterly review that a goal quietly died in week three. When your attention is spread across more output than ever – some of it produced by agents – this is the early-warning system for the work that actually matters. Goals rarely fail loudly. They slip quietly, one missed update at a time.
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.