Product · Wednesday, 26 August 2026 · Issue 86 · 10 stories

The product problems you can't A/B test

Waymo's product chief ran a ten-year program with no scorecard. Most product managers now face that same impossible measurement problem.

Product Management

Roadmaps fail on missing steps, not bad estimates, hunt the precondition early

A 60-engineer SaaS company hit every estimate yet shipped seven months late: its identity vendor wouldn't sign a regional agreement inside 90 days, killing the deadline in January without anyone noticing. Estimation error is bounded and can be buffered; a missing step can only add time. Work backwards from what must be true before execution, and get that answer before you schedule it. The fix is planning what isn't on the list yet.

codegood.co ↗11 min read at source

Waymo Spent a Decade on a Problem You Can't A/B Test. That's Most PM Jobs Now

Waymo's product chief ran a ten-year program with no scorecard, no playbook, and nothing to split-test. His escape: define what good actually looks like, lock in concrete checkpoints, and build a team that can learn its way forward. As software moves into physical and ambiguous problems, that job is becoming every PM's job.

theskip.substack.com ↗11 min read at source

A Good Product Move Opens Future Options, Not Just Today's Win

Wayward product teams, unlike wayward hikers, don't turn back at a dead-end fork: they camp, hack through brush, ship the quarter, and quietly retrace their steps while no one discusses it. Dave Evans argues each move should be "enabling" as well as bounded, grounded, and constructive. A move that's only valuable in isolation is cheap; one that sets up the next game is strategy. The lesson is to weigh every ship against the options it leaves behind.

davesresearch.com ↗5 min read at source

Design & UX

Stop Rebuilding Legacy from Scratch: 65-79% Fail. Modernize in Slices

Legacy is rarely broken; it's misaligned and business-critical, which is exactly why a big-bang relaunch is the riskiest path. Up to 79% of modernization projects fail or get cancelled, often because hidden dependencies surface too late. The play: replace workflows in small slices, keep users in the loop, and deliver value early. Slow and strategic beats dramatic and catastrophic.

smart-interface-design-patterns.com ↗9 min read at source

Don't hand UX research to AI: keep the researcher in the loop

Most AI-enabled UX research still casts the tool as a black box that spits out findings. Jennifer Bowie proposes the opposite: a researcher-in-the-loop model where the human sets questions, governs the AI's reasoning, and owns every conclusion. It's a concrete counterweight to the "AI does your research" hype, and a framework teams can actually adopt before automation erodes the craft.

uxdesign.cc ↗10 min read at source

Strategy & Growth

Your Site Is Best the Day It Ships. It Doesn't Have to Decay After.

A website is perfect for one day, launch day, then quietly drifts into a period piece as markets and messaging move. Pierre Burgy argues agents should keep improving it after it ships, but the real lesson from building the autonomous-platform Fimo is that nobody wants full autonomy. The work splits into a large no-brainer pile (rule-bound fixes) and a small human-judgment pile, and the line between them matters more than the tech.

smashingmagazine.com ↗8 min read at source

Netflix may sell rivals' streams, a smart move that betrays its ambitions

Stratechery's take: Netflix is reportedly considering reselling other streaming services, pivoting from a library owner to a marketplace aggregator. Thompson calls the idea smart, yet a let-down for the original goals and bigger pivots Netflix once hinted at. Where it wins resembles Roku: owning the customer relationship while the content's fate stays out of its hands.

stratechery.com ↗3 min read at source

Healthcare AI's Oracle Problem: benchmarks can't prove that a model helps real patients

Scoring 100% on med-school tests doesn't make a doctor (or an AI) good at real patients, and that gap is the invisible bottleneck to healthcare AI. Evals in medicine are quietly subjective, trained on what physicians happened to do that day, and the data behind them is guarded and easily spun into marketing. The only honest measure requires tracking a patient forward over time, which is far harder than passing a benchmark. Buyers of hospital AI have no clean way to tell who's best.

a16z.news ↗13 min read at source

Tools & Launches

Spinifex re-implements the AWS API surface so you can run "AWS" on your own hardware

Spinifex is an open-source platform that recreates EC2, EBS, S3, VPC, IAM, EKS and more on bare-metal, edge, or on-prem hardware, with the AWS CLI, SDKs, and Terraform working unchanged. No rewrites, just redeploys. That opens the door to air-gapped cloud, cut costs, and data you fully control, without the hyperscaler lock-in.

github.com ↗8 min read at source

13 hands-on Docker security labs, from CIS checks to AI context poisoning

A free repo of reproducible Docker security labs that move from docker-bench CIS auditing and least-privilege hardening through secrets, Cosign image signing, and Kyverno policy gates to AI context-poisoning defenses. Each lab ships scripts, configs, and captured experiment evidence rather than theory, and maps to production cloud/K8s ops. Solid for engineers, auditors, and platform teams wanting something to actually run.

github.com ↗10 min read at source

You’re caught up.

That was all 10 stories in the 26 August issue.