You're automating the wrong noun
The biggest gains do not come from automating isolated tasks; they come from redesigning the workflows around them.
Observe a software org that's gotten real speed improvements from AI and the surprising part isn't the tooling. It's the operating model: fewer handoffs, blurrier roles, workflows that look nothing like last year's process diagram. Task-level automation alone doesn't get you there. You can draft every report and triage every ticket with an agent and the org will still move at its old speed, because the speed was never in the tasks; it's in the workflow wrapped around them.
Two things that showed up in this week's readings that we see consistently in orgs getting the most out of AI: they're automating workflows, not tasks; and they're not afraid to let roles blur.
Here's what we're reading this week.
Uber's Agentic Pods
Uber's CTO Praveen Neppalli posted the company's adoption numbers: 99% of engineers on AI tools, more than 70% of PRs attributed to agents, 2,500+ agent skills across the SDLC. Headline numbers, but the real takeaways are the question Uber asked next (how do we scale this beyond engineering?) and the structure they built to answer it. They paired 30 of their most AI-fluent engineers with domain experts from other functions (finance, legal, ops, support) and gave each pod ten days: two to shadow the expert, one to prioritize, two to build a working agent alongside the person doing the job, four to validate that it generalizes, one to ship. Sixteen pods across sixteen functions in two months, with results like capital allocation across 150 cities going from 15 hours to 30 minutes.
The workflow becomes the unit of automation - not the individual task.
Automate the task and you save minutes; redesign the workflow and the whole org gets faster.
Figma's 2026 AI Report
Figma's 2026 AI Report reinforces some of the same ideas at a much wider scale (8,403 survey responses, 639 interviews, ten markets). The stat to watch is the role blur. Developers participating in design went from 44% to 60% year over year, designers participating in development doubled from 21% to 41%, and 41% of respondents now say AI meaningfully changes how their teams work together, up from 7% two years ago.
Redesign a workflow around an agent and the handoffs are the first thing to go; when the handoffs go, the role boundaries drawn around them go too. Role blur is a symptom of workflow redesign, not a fad, and the trend line says it's accelerating. Or as the report puts it:
When AI can 10x anyone, personal productivity stops being the thing to optimize for.
Product Design is Dead; Long Live Product Design
John Smothers argues that every craft gets reborn when its practitioners finally touch the real medium instead of an artifact one step removed: musicians moved from tape to the DAW, film editors from celluloid to the timeline, and architects traded the drafting table for CAD. Designers, he says, are next, because the medium of software is code, and everything produced in a design tool was only ever "a picture of what could be."
We called it 'handoff' and built an entire professional culture around managing this inefficient and lossy workflow.
One thing to try this week
A one-day version of Uber's pod. No budget, no 30-person program, no exec sponsor required.
Pick one workflow your team keeps "automating" a task at a time. Sit your most AI-fluent engineer (that might be you!) next to the person who actually does the work. Don't build anything yet; shadow the workflow end to end for a day and write down every handoff, approval, and tool switch. Then ask one question: if we redesigned this around an agent, which of these steps stop existing? That list is your real backlog.
If the answer turns out to be "most of them," congratulations, you've found your first pod.
Happy shadowing. It's more fun than it sounds, we promise.
— Claire + Zach