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.
Four hours a week
Most managers who “used to code” pick one of two bad options: stay hands-off and lose the ground truth, or wedge themselves back into the sprint and become the bottleneck they used to complain about. Karol Wojciszko argues for a third path: carve out roughly 5-10% of your week, about four hours, for code that has no deadline, touches one service, and blocks nobody. The point isn’t shipping. It’s staying fluent in what your team actually deals with day to day: the flaky test, the migration nobody wants to touch, the codebase’s quiet weirdness. His approach is to learn, explore, plan, self-review, verify, then submit the PR through normal team review like anyone else. No special treatment for the boss’s code. He’s candid that AI tools change the math here too: use a cheap model to learn the terrain, a stronger one to plan, and start fresh sessions before context rot sets in. The real deliverable here is a manager who still knows what “done” costs, and a team that trusts him that much more.
— Karol Wojciszko, 6m, #engineering, #leadership, #management
Polish isn’t proof
AI-generated code looks done – clean formatting, docstrings, passing tests – and that’s exactly the problem. Codacy’s engineering team makes the case that “looks production-ready” and “is production-ready” have quietly become two different claims. The data backs it up: as teams move from low to high AI adoption, the incidents-per-pull-request ratio jumps 242.7%, and unreviewed merges climb right alongside it. Worse, developers using AI tools believe they’re moving faster even when they objectively aren’t, which means the people best positioned to catch a problem are the least likely to look for one. Their fix is deterministic gates: baseline scans that run on every AI-assisted PR regardless of who wrote it, human review mandatory for high-risk changes, and AI review kept strictly advisory.
— Codacy, 7m, #ai, #engineering, #security
Indirect agents
Forget the debate over whether your company should build its own AI agents. Isaac Sacolick’s read is that the decision is already being made for you, by whoever sells you SaaS. Enterprise platforms are embedding agents directly into the tools you already run, and the strategic question isn’t build-versus-buy anymore, it’s which vision of automation you’re inheriting. SAP is racing toward a fully “autonomous enterprise” – its Joule Agents jumped from 40 to over 200 in a year. Atlassian is betting on augmentation, AI that sharpens human judgment rather than replacing it. Others sit in an uneasy human-in-the-middle phase, using oversight to build trust before handing over more control. Sacolick’s advice for CIOs: push vendors for transparency about what their agents actually do, watch for shadow AI creeping in through tools you didn’t vet, and remember that every agent you adopt without asking is still a choice, just one somebody else made for you.
— CIO, 8m, #ai, #agents, #strategy
Wrong and right
Jeremy Theocharis doesn’t argue with the LLM skeptics – he agrees with their arguments. Training data built on scraped copyrighted work, real environmental cost, junior engineers losing the base work that used to teach them the craft, and open source drowning in low-effort AI slop contributions. But he also spent $9,838 on tokens in a single month and isn’t slowing down. His perspective is simply that LLMs amplify whatever thinking you bring to them. Hand one a clear framework and it supercharges you; hand one nothing and it hands back fluent slop, dressed up to look like insight. His filter for when output is safe to trust: would you stand in front of an audience and read this aloud? Hesitate, and it’s slop. He only delegates work where he already knows what good looks like, which is a smaller circle than most power users admit. The takeaway isn’t that the critics are wrong. It’s that being right about the problems doesn’t obligate you to opt out of the tool – it just obligates you to use it with eyes open.
— Jeremy Theocharis, 9m, #ai, #engineering, #productivity
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.
The conversation you’re avoiding names the overdue talk you keep bumping to next week — the disengaged teammate, the blocker that never actually resolves, the goal that’s quietly lost its pulse. Every Friday at 3 PM it scans your recent check-ins, blockers, and goal updates for the patterns that spell avoidance, then names the single conversation that matters most and suggests how to open it. Not a to-do list — one specific nudge, timed so you can’t carry it into the weekend. Hard conversations rarely fix themselves by waiting. This one just makes sure you can’t say you didn’t know.
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.