The best of AI and product news for PMs, curated and contextualized so you can spend less time scrolling and more time building.

• OpenAI's GPT-5.6 goes public after a monthlong government standoff. GPT-5.6 first showed up in late June, but only as a locked-down preview available to about 20 partners whose names OpenAI cleared with Washington one by one, a first for an American frontier model. That hold lifted yesterday, when all three tiers went public across the API, Codex, and ChatGPT, Sol for flagship reasoning, Terra at roughly half the cost of GPT-5.5, and Luna for fast, cheap volume. The real story for product teams isn't a stronger model, it's the precedent, because the most capable system you build on now ships on a clock Washington helps set.
  
• Claude Cowork lands on web and mobile. Anthropic brought Cowork to web and mobile with sessions that sync across devices, so a task you hand off follows you from laptop to phone. It's worth asking who your next user really is, because when people delegate work to an agent that moves across apps, your product increasingly has to serve the agent doing the task, not just the human clicking through the screens.
  
• Cloudflare moves to block AI crawlers by default. Cloudflare now lets any site sort AI traffic into three buckets, Search, Agent, and Training, and from September 15 it will block the Training and Agent crawlers by default on ad-supported pages for new and free-tier sites, while letting search crawlers through. Product teams whose reach depends on open-web data, or whose features send agents out to browse, should start planning for a web where crawling is opt-in and increasingly paid.
  
• The White House is finalizing voluntary AI release standards. Following June's executive order, the White House is racing to finalize voluntary standards with OpenAI, Google, and Anthropic that set benchmarks and timelines for testing frontier models before launch, the same review GPT-5.6 just cleared. This one matters most to teams whose plans assume the next model shows up on schedule, since the framework is starting to decide who gets the most capable systems, and when.

Lenny Rachitsky and Noam Segal ran their second annual survey of how tech workers feel, and the headline finding is a fault line. One half feels amplified by AI, more capable and confident than they've ever been. The other half feels shaken by it, less sure there's still a place for them. Which side you land on now predicts how you feel about your career more than role, seniority, or company size, even as significant burnout climbed to 55.7% and career optimism slipped below half.

For product leaders, that split is a management problem you already own, not just a survey stat. The people on your team aren't divided by which tools they were handed. They're divided by whether the work still feels like theirs or feels automated out from under them, and that framing is yours to set. Take that into your next round of one-on-ones. The line isn't AI haves versus have-nots. It's who has been shown their judgment still matters, and who has been left to assume it doesn't.

Ant Murphy, product coach and founder of Product Pathways, tells a story about a PM whose founder had crowned monthly active users the North Star. Murphy asked what monthly behavior would actually make MAU worth chasing, and the honest answer was none. They were an insurance startup, and a customer opening the app every other day usually means something has gone wrong. His test for any metric is two questions. What are you trying to learn, and what will you change as a result?

For product leaders, that's a cheap audit to run before the next planning cycle. For every metric on the dashboard, name the decision it drives and the number that would trigger it, then cut the ones where the honest answer is "nothing." Murphy's harder point is that the metric that matters is often the one that's annoying to measure. A meaningful proxy beats a generic figure pulled off the shelf, even when it takes more digging to define.

When you adopt an AI product tool, what actually changes about your day? Tim Herbig, a product coach, put that to Jason Kothary, a solo PM at a Canadian nonprofit. The speed was real, roughly 40% faster user-story drafting and about a quarter less time on backlog refinement. The more telling part was what Kothary did with the time he saved. Rather than ship faster, he added a step that hadn't existed before, pressure-testing each idea against the tool as a "validation sparring partner" before any story earned its place.

For Herbig, the PMs who get the most from these tools won't be the ones who generate the most, but the ones who put the saved capacity back into the work the tool can't do, validating harder and staying close to the users it will never talk to. That's the practical version of Lenny's "amplified" camp. Let the tool hold your context and push back on your thinking, then spend the freed hours on the calls only you can make. The tool in Kothary's case is Productboard's Spark, and we're grateful Tim gave it an honest look. For the record, we didn't pay for or sponsor the piece in any way, every word of it is Jason's and Tim's.

FROM THE FEED

Judgment over volume

• Shreyas Doshi makes the case that product sense is the one skill AI can't hand you. When anyone can spin up forty versions of a feature, the value is in choosing the right one. (Read his take).

• Lenny Rachitsky reshared Elena Verna's "AI Confidence Theater" piece and admitted it made him feel calmer, which will land for a lot of us. Verna's push is to stop performing AI mastery you don't have and just show the real workflow. (Read the post).

• Jordan Nolff demoed Spark live with Productboard CEO Hubert Palan on their own data, running the full loop from idea to spec to shipped to measured impact. (Watch the demo).

FROM PRODUCTBOARD

Put AI to work on your own product, live this July

Spark Forge: Product Teams Series | Every Thursday in July · 45 min · virtual

Spark Forge is a run of hands-on working sessions where you build on your own product with live facilitation from core Productboard team members. Each Thursday takes on one part of the Spark workflow. July 16 covers surfacing the signal that matters, July 23 going from signal to a delivery-ready spec, and July 30 measuring what shipped and closing the loop. Pick the sessions you want at signup, and move quickly, since seats are limited.

Save your seat →

Prefer to start now? Try Spark for free →