We track Product so you don't have to. Top Podcasts summarised, the latest AI tools, plus research and news in a 5 min digest.
Hey Product Fans!
Welcome to this week’s 🌮 Product Tapas.
New here? We're your shortcut to staying sharp. Essential stories, practical tools, real insights.
For the best reading experience, check our web or app version and sign up for future editions here.
What’s cooking this week? 🥘
The biggest story barely made the headlines it deserved: a US directive forced Anthropic to switch off Fable 5 and Mythos 5 for everyone, overnight. Around it, Apple shipped a Siri that wins on "good enough, by default" rather than best-in-class, the model price war turned explicit, and agents moved from party tricks to plumbing. Plenty to chew on.
📰 Not Boring -> Anthropic pulling Fable 5 and Mythos 5 on a government order, Apple's default-everywhere AI bet, the model price war turning explicit, the open web getting squeezed, and the product strategy playing out in the margins
⌚️ Productivity Tapas -> offline meeting transcription you can flag mid-call, an open-source UI kit for document apps, CLI-controlled VMs, and a tool to fight AI skill rot
🍔 Blog Bites -> the tiny tells that give away AI-built design, why AI made good recruiters better not redundant, and why Atlassian bought a startup it didn't really want
🎙️ Pod Shots -> The Human Edge - four leaders on the one thing AI still can't do for you: influence and judgement
Let's go 🚀
📰 Not boring
🍎 Apple Ships Its AI Bet
Apple's New Siri Is Just Good Enough to Ease Its AI Crisis and Apple Wins Consumer AI By Default - the new Siri is not class-leading, but it is on a billion devices by default, which is its own kind of moat
Apple Foundation Models and Apple's open-source container project - Apple shipped on-device model access and developer tooling, making AI a first-class platform capability rather than a feature
Android 17 launches with new multitasking tools as Google expands Gemini features - the other side of the duopoly answered the same week, baking Gemini deeper into the OS layer
The iPhone's Last Stand and WWDC26 - The Small Things - the strategic read on whether ambient AI erodes the device that made Apple the most valuable company on earth
Apple is betting that "good enough, everywhere, by default" beats "best, somewhere, opt-in" - the same distribution logic that won it the App Store. For PMs, the lesson is that frontier capability is rarely the thing that wins a platform; the install base and the default setting usually do.
🤖 The Model Layer Keeps Commoditising
Initial impressions of Claude Fable 5 - Anthropic's new top tier matches rivals on capability but ships at twice the Opus price, betting that buyers pay for judgement and guardrails, not raw benchmarks
OpenAI Considers Drastic Price Cuts, Anticipating War for Users With Anthropic - the price war the Fable 5 launch implied is now explicit, and it is a race to the bottom on the model itself
Running local models is good now and Apple Foundation Models - capable models now run on a laptop or a phone, which removes the API bill from a growing slice of workloads
ChatGPT's market share slips below 50% for first time - the first-mover's dominance is eroding as the field fills with credible substitutes
Brian Armstrong's take that 80% of workloads run on 99% cheaper models within 12-18 months and OpenRouter's live model rankings - the market is already routing most volume to the cheapest model that clears the bar
Every signal this week points the same way: capability is converging, price is the battlefield, and a lot of work is moving to models that cost almost nothing. The PM discipline that matters now is matching model intelligence to task complexity - defaulting to a cheap model and reaching for the expensive one only on the calls that genuinely need it.
🧠 Agents Move From Demo to Workflow
Coinbase's new tool lets agents trade and pay for premium research - an MCP server that gives agents a real payment and execution surface, not just a chat box
DoorDash's new AI chatbot lets you order with prompts and photos and Meta's Edits app is getting an AI assistant and a desktop version - consumer products are embedding agents into the core flow rather than bolting on a separate assistant
How Coinbase built an AI agent that converts Figma designs into production-ready code and The PM's playbook for shipping AI features that actually work in production - two practical guides on getting agentic features past the demo stage and into shipping software
Announcing Stack Overflow for Agents and Building a Good Vertical Agent - the infrastructure and patterns for agents are maturing into named, reusable building blocks
Agentic Code Review - Addy Osmani on handing the first pass of code review to agents, a concrete production workflow rather than a thought experiment
The agent story this quarter is no longer "can it hold a conversation" - it is "where does it sit in the workflow and what can it execute." Coinbase giving agents payment rails and DoorDash putting one in the ordering flow are the same move: the agent becomes plumbing, not a feature. PMs should be mapping which step in their product an agent removes, not whether to add a chatbot.
🌐 The Open Web Gets Squeezed
The Web We Know Is Going to Disappear - a long argument that AI answer engines are quietly dismantling the link economy the web was built on
Google Chrome's next update will mark the end of popular ad blockers and Google Chrome is killing all uBlock Origin bypasses, with Edge and Opera to follow - the browser layer is closing the door on user-side control of what loads
Meta's new 'AI Mode' on Facebook pulls from public info across its platforms - the platforms are turning their own user-generated content into AI training and answer surfaces
Deezer's new tool can identify AI music from Spotify, Apple Music, and others - as synthetic content floods the pipes, provenance detection becomes a product category of its own
Why developers use LLMs to write blog posts - even the people who make the web are now generating it with models, which compounds the provenance problem
The same week Chrome tightens its grip on what users can block, AI answer engines are eroding the reason anyone clicks through to a site at all. For PMs who depend on search and content distribution, the ground is shifting twice over - the browser is getting less open and the destination is getting less visited. Provenance and direct relationships look like the hedge.
UK bans under-16s from using social media apps including TikTok and YouTube and UK unveils sweeping social media ban for users under 16 - a major market just removed a whole age cohort from the addressable base of every social product
These are the countries moving to ban social media for children - the UK is not an outlier; age-gating is becoming a global regulatory pattern product teams have to design for
Trump signs AI executive order to increase government oversight and For a Second Time, Trump Muses About Americans Sharing in AI Wealth - US AI policy is moving from voluntary commitments toward executive-level oversight and redistribution talk
India orders temporary ban on Telegram over exam fraud concerns - a reminder that platform access in large markets can vanish overnight on a regulator's call
The free-build era for consumer platforms is closing. A UK under-16 ban is not a content-moderation tweak; it is age verification, identity, and a structurally smaller market baked into the product roadmap. Any PM shipping a social or UGC product now has compliance as a first-class design constraint, not a legal afterthought.
🚫 The Model That Got Pulled
Anthropic disabled Fable 5 and Mythos 5 for everyone after the US ordered it to bar foreign-national access and the Trump administration cited national security - rather than build a way to fence off foreign users, Anthropic pulled its two most capable models for all customers to comply
The directive reportedly followed a jailbreak found in Fable 5 and Anthropic disputes the standard - the company warns that applied consistently it would halt all new frontier-model deployments across the industry
This is the story every PM should sit with. A government directive took the most capable model on the market offline for everyone, overnight, with no warning. If your product depends on a frontier model, regulatory access risk just moved from a hypothetical to a real entry on your risk register - and the mitigation, model-agnostic plumbing that can fail over to another provider, is exactly the harness pattern the rest of the industry is converging on anyway.
📱 Product Strategy in the Margins
Threads adds new personalization and community features as it reaches 500M monthly users and Bluesky launches group chats as it shifts focus to community features - the Twitter-replacement field is converging on the same answer: retention lives in community and DMs, not the public feed
Pool's new app turns your screenshots into a searchable memory bank - a tidy example of finding product in a behaviour people already have rather than inventing a new one
Waymo launches premier subscription tier for $29.99 a month - autonomous mobility is now experimenting with tiered subscription monetisation, the most ordinary product move imaginable for a futuristic service
Software Is Not A Single-Player Game and Doing nothing at work - two sharp reads on collaboration and slack as features of how software actually gets built
Threads and Bluesky reaching for community and group chat in the same week is the clearest signal that the public-feed model has hit its retention ceiling. The pattern worth lifting is Waymo's: even the most novel product eventually wins or loses on boring fundamentals like tiering and pricing. Novelty gets you launched; the ordinary moves keep you alive.
⌚ Productivity Tapas: Time-Saving Tools & Workflow Automation
Trace: offline Mac meeting transcription that lets you flag moments mid-call so you can jump straight to the bit that mattered. Fully local, so nothing leaves your machine - useful for PMs in sensitive customer or strategy calls.
Extend UI: open-source UI kit purpose-built for modern document apps - editors, viewers, and AI-document interfaces. A fast head start for any team building a document-heavy product without reinventing the components.
machine0: persistent NixOS virtual machines you spin up and control entirely from the CLI. Reproducible, disposable environments for running agents or test workloads without touching your main setup.
Fata: spaced-repetition practice designed to fight skill rot from over-relying on AI coding. Keeps your fundamentals sharp when the agent is doing most of the typing - a neat counter to the deskilling worry running through this week's reads.
Homebrew 6.0.0: the big release of the macOS package manager every developer already uses, with a faster install path and cleaner dependency handling. Worth the upgrade if you live in the terminal.
Remember. Product Tapas subscribers get our complete toolkit - 550+ personally tailored, time-saving tools for PMs and founders. Your shortcut to efficiency and what's hot in product management 🔥
Check the link here to access.
🍔 Blog Bites - Essential Reads for Product Teams
Design & UX: The Tiny Tells That Give Away AI-Built Design
Dan and Louis-Xavier at Growth.Design pull apart a music app interface to show the small but giveaway signs that AI built it - and why design fundamentals still decide the result. Read the full article here.
💡 "AI can't really 'see', so it builds things without knowledge of how it would be experienced in reality."
| | Key Takeaways
AI misses the tiny details : Because it has no human eyes, AI gets the small experiential details wrong - things a designer would catch instinctively because they imagine how the screen is actually used
It breaks the proximity principle : In the example, a playlist dropdown sat directly below the TV controls, so it looked like it would change the channel. The Gestalt proximity principle says elements placed close together appear related - and AI repeatedly violates it
The fix is grouping and spacing : Group similar actions together and create more space between distinct sets of actions. Small layout moves, large clarity gains
AI is the accelerator, not the designer : The person who understands design fundamentals and the psychology of why something works will get far better results from the same AI tool
Dan and Louis-Xavier, Growth.Design
Why I picked this: A concrete reminder for the vibe-coding era that runs right through this week's model-commoditisation theme: AI gets you running fast, but the human who knows why a layout works still owns the quality bar before anything ships to users.
People: AI Made the Good Recruiters Better, Not Redundant
Bill Kerr interviews Sofia Faiman, Recruitment Manager at remote talent platform Athyna, on what actually changed when her team moved from spreadsheets to an AI-driven hiring process - and why empathy, not automation, stayed at the centre. Read the full article here.
💡 "At the end of the day, AI didn't replace the recruiter. It made the good ones much better."
| | Key Takeaways
Automate the mechanics, protect the relationship : Faiman's team automated job descriptions, interview notes, briefings and language scoring, freeing recruiters to focus on relationships, strategy and the decisions that genuinely need a human
The numbers moved : AI-powered matching against client briefs improved endorsement times by 50%, alongside better match quality and faster time-to-place. An AI-led interview now self-serves in around eight minutes
Candidate experience is the thing that breaks : The core dysfunction is treating people as interchangeable resources. Transparent, consistent communication - even when things slow down - is her antidote
Lead the AI transition deliberately : Facing real fear in the team, she ran AI literacy sessions so nobody felt left behind - a process that spawned an entirely new business unit
Bill Kerr, Open Source CEO
Why I picked this: The "AI made the good ones better" framing is the honest version of what augmentation looks like in any function, and it pairs neatly with the agents-into-workflow theme above. I picked it for the rare combination of concrete metrics and a clear stance that empathy and AI accelerate each other rather than compete.
Case Studies: Why Atlassian Bought a Startup With a "Rounding Error" Customer Base
Tom from Strategy Breakdowns interviews Mehdi Boudoukhane, former founder of customer-feedback startup Cycle and now Principal PM for Feedback at Atlassian, on the real acquisition story, Atlassian's first swing at usage-based pricing, and the brutal distribution maths of selling into a 300,000+ customer base. Read the full article here.
💡 "In an acquisition you can buy 3 things: the business (customers, ARR), the product, and the team. In this case Atlassian didn't want the business. It was too small. It was really the other 2."
| | Key Takeaways
Acqui-hire for know-how : Cycle had a few hundred customers against Jira Product Discovery's 20k+. Atlassian bought a six-year head start and a team that had been processing millions of feedback items a month - JPD had not shipped a single AI feature before they joined
Kill the legacy product : They sunsetted Cycle entirely. The learning from past acquisitions: keep the legacy product and you maintain two products at once, when the goal was getting from 20k to 100k customers
Autopilot, not copilot : The Feedback App processes every item automatically rather than waiting to be prompted, which makes token costs and revenue predictable and underpins Atlassian's experiment with usage-based AI pricing where a 100x gap between token cost and customer value leaves room for margin
The distribution gift : A 15-minute form sent only to JPD users drew 7,000 teams asking to talk in a few days; they ran 100 customer interviews a week, where as a startup Cycle had struggled to book five or six
Tom, Strategy Breakdowns
Why I picked this: A rare, candid founder's-eye view of what acquirers actually buy and what they throw away. The autopilot-versus-copilot framing for predictable AI pricing is a genuinely useful mental model for anyone wrestling with how to charge for AI features - which ties straight back to this week's model-pricing theme.
🎙 Pod Shots - Bitesized Podcast Summaries
Remember, we've built an ever-growing library of our top podcast summaries (120 or so). Whether you need a quick refresher, want to preview an episode, or need to get up to speed fast - we've got you covered.
Check it out here
🤖 The 4 Product Leadership Skills AI Still Can't Touch
The skills that get you promoted now are the ones a model can't do for you. Across four recent podcast conversations, leaders from Webflow, Harvey, a fractional CPO and a public-sector veteran kept circling the same point: as AI absorbs the production work, judgement, influence and the courage to make hard calls become the whole game..
Key insights from the full article:
🎯 Influence is not politics - Jessica Fain reframes the skill people find "icky": politics manipulates outcomes for your gain; influence just increases the odds your good ideas survive.
🧠 Treat your exec like your hardest user - The curiosity and empathy you bring to customers vanishes the moment you walk into a leadership meeting. Bring it back, and treat the conversation as discovery.
🎁 Show options, not your homework - How much to reveal, when to bring three choices, and why the proof belongs in the appendix.
🗣️ Lead with the recommendation, hold the detail - Dave Martin's "signal prep" and "CALM" framework turn chronic over-explainers into people who lead the room.
💔 A people leader's job isn't to make you happy - Katie Burke on building the table instead of asking for a seat, and why the hardest calls are the most human.
🪑 The resort has to match the brochure - There's no globally correct culture, only an honest one. Protein versus sugar, and the "Berrygate" lesson.
🗑️ Unbuilding is the answer - Ayushi Roy on why the senior move is often deleting ten systems instead of shipping an eleventh.
🤝 AI is a spell-checker for influence - The tooling doesn't do the judgement; it lowers the cost of communicating and producing it. The thinking still has to be yours.
That’s a wrap.
As always, the journey doesn't end here!
Please share and let us know what you would like to see more or less of so we can continue to improve your Product Tapas. 🚀👋
Alastair 🍽️.