
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 IPO floodgates creaked open - Anthropic filed to go public, nudging a trillion-dollar valuation (because why not), and the question now is just who prints a number first. Microsoft quietly decided its biggest AI supplier should also be its competitor and shipped its own models to cut the OpenAI bill. And in the story I can't stop thinking about, Karpathy's Autoresearch agent found a three-year-old bug in PostHog's production code and made it 11% faster - turns out agents are more useful when they fix things than when they chat. Meanwhile everyone with a hardware division is betting your next computer is something you wear on your face, and the entry-level job market is quietly repricing around a skill - judgment - that internships used to exist to teach.
📰 Not Boring -> the IPO floodgates, the model layer going four-way, agents shipping real fixes, the wearable AI bet, and a repricing job market
⌚️ Productivity Tapas -> Mina (an AI teammate that joins your calls), Tabstack's cited-research API, Typeahead, Databox MCP, and Department of Product's prompt library
🍔 Blog Bites -> the new types of AI-driven tech debt, Figma's CEO on the "design vs code" false dichotomy, and how Dropbox built its Nova coding agent
🎙️ Pod Shots -> #138 "We're in 1997" - Benedict Evans, a16z, and Dax Raad on the gap between AI hype and the AI you're actually working in
Let's go 🚀
📰 Not boring
💰 The IPO Floodgates Open
Anthropic files to go public and nears a $1 trillion valuation - last week it was "preparing to file", this week the S-1 is real [UPDATE]
Why it matters if OpenAI or Anthropic wins the IPO race - the first lab to the public markets sets the comparable everyone else gets priced against
The SpaceX IPO and data centres in space - Thompson on why orbital compute is the real story behind the filing
IBM commits $10bn to quantum computing - the capital is now flowing into bets that pay off in 2029, not next quarter
The AI labs are racing to the public markets while they can still command private-market multiples. For PMs, the signal is in the comparables: once Anthropic or OpenAI prints a number, every AI feature in your roadmap gets benchmarked against a trillion-dollar reference point.
🤖 The Model Layer Stops Being a Two-Horse Race
Microsoft unveils its own AI models to cut its reliance on OpenAI and lower developer costs - the most important customer is building the alternative
Microsoft and Google are late to AI coding but "absolutely critical" they compete - the coding-model market just went from two players to four
Claude Opus 4.8 lands as "a modest but tangible improvement" - Willison's read on where the frontier actually is
AI models are having their iPhone moment - Malik on what comes after the capability race
When your biggest cloud reseller starts shipping its own models to reduce dependence on you, the moat was never the model. PMs picking an AI provider this year are buying into a market that's commoditising faster than the pricing pages admit. Although, Claude is still my fave until proven otherwise.
👨💻 Agents Move Into the Workflow
Introducing Dynamic Workflows in Claude Code - orchestration moves from prompt-chaining to declared, repeatable flows
Microsoft launches AI that works like an executive assistant and Inside Microsoft's Project Solara - the assistant pattern goes from demo to product
Karpathy's Autoresearch found a 3-year-old bug in PostHog's query engine and improved performance 11% - agents doing real maintenance work, not toy tasks
Rethinking search as code generation - Perplexity reframes retrieval as a code-gen problem
The interesting agent stories this week aren't launches, they're outcomes: a real bug found in production code, a real performance win. The bar for "useful agent" just moved from "can hold a conversation" to "can ship a fix".
🥽 The Hardware Bet on Ambient AI
Apple seeks to disrupt the glasses market the way it did watches, while iOS 27 overhauls Siri - the long game on a face-worn assistant
Nvidia introduces the first PCs designed for AI agents and Nvidia-powered Windows 11 PCs leak - the hardware is being rebuilt around agents, not apps
Meta is working on an AI pendant and more smart glasses - the wearable-AI form-factor war widens
Intel says its upcoming AI chip will be cheaper and cooler than Nvidia and AMD - a price-led challenger enters
Every major platform is now betting that the next computer is something you wear, not something you open. PMs building consumer products should watch which input model wins - voice, glance, or gesture - because it changes what "an interface" even means.
📉 The Job Market Reprices Around AI
How one tech company created 13 new types of jobs because of AI - the roles AI creates, not just the ones it removes
AI is killing the summer internship and remote work - not AI - has sidelined recent graduates, research finds - two competing explanations for the broken entry-level pipeline
Judgment is the skill that matters most in the AI era - when execution is cheap, knowing what to build is the differentiator
The best engineers write less code - output stops being the measure of a good builder
The entry-level pipeline that trained a generation of PMs and engineers is genuinely breaking, and the data disagrees on why. The skill the market is repricing toward - judgment over output - is exactly the thing junior roles used to exist to build.
Join 2M+ Professionals Getting Ahead on AI
Keeping up with AI shouldn't feel like a second job.
But between the new tools, viral posts, and endless hot takes, most people spend hours every week trying to figure out what actually matters.
The Rundown AI fixes that.
It's a free newsletter that gives you the AI news, tools, and tutorials you actually need to know. All in just 5 minutes a day.
Over 2M professionals at companies like Apple, Google, and NASA already read it every morning to stay ahead.
Plus, if you complete the quiz after signing up, they'll recommend the best tools, guides, and courses for your specific job and needs.
⌚ Productivity Tapas: Time-Saving Tools & Workflow Automation
Mina Meeting Assistant: an AI teammate that actively participates in calls - speaks, pulls context from 200+ tools (Slack, Jira, Notion, HubSpot), and executes tasks while the conversation is still happening. For PMs who lose half a meeting capturing actions.
Tabstack Research: Mozilla-backed web-research API that returns cited answers from the live web in one call, every claim backed by a source URL. A clean way to ground AI features in verifiable research instead of hallucinated summaries.
Typeahead: a local, offline AI writing assistant giving inline suggestions across every Mac text field - privacy-conscious autocomplete that works in your PRD tool, your email, and your tickets without sending text to the cloud.
Databox MCP: connect your BI dashboards to Claude and ChatGPT so you can ask plain-English questions of your product metrics and get answers in the chat, not a spreadsheet export. Included free for all Databox users.
Department of Product's AI Prompt & Skills Library: a growing, curated library of prompts and Claude Skills built specifically for designers and product managers - a shortcut to not reinventing the same prompt every week.
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

🍔 Blog Bites - Essential Reads for Product Teams
AI/Tech: The 4 New Types of AI-Driven Tech Debt
AI-assisted development is shipping code faster than ever - and quietly creating categories of technical debt that didn't exist two years ago. This piece breaks down the new failure modes (prompt debt, retrieval debt, evaluation debt) and how to prevent each before it compounds. Read the full article here.
💡 "The cost of AI-generated code isn't writing it - it's the debt nobody decided to take on."
| | Key Takeaways
Comprehension debt : code that works but no one on the team actually understands, making every future change riskier
Prompt debt : workflows built on opaque prompts that break silently when the underlying model updates
Review debt : volume of AI output outpacing the team's ability to meaningfully review it
Convention debt : AI tools generating inconsistent patterns that erode the codebase's internal coherence
VentureBeat
Why I picked this: It connects directly to this week's agents-in-the-workflow theme - the same tools letting agents ship real fixes are the ones quietly accumulating debt no PM has put on the roadmap.
Strategy: Figma's CEO on Why "Design vs Code" Is a False Dichotomy
As AI collapses the gap between designing and building, Figma's CEO argues the old turf war between designers and engineers is the wrong frame entirely. A sharp read on where product craft is heading when the tools stop respecting the boundaries. Read the full article here.
💡 "The question isn't whether designers should code - it's why we ever drew the line there."
| | Key Takeaways
The boundary was a tooling artefact : design and code were separate because the tools were separate, not because the work is
AI dissolves the handoff : when the same prompt can produce a mockup or a working component, the handoff stops being a stage
Craft moves up a level : the differentiated skill becomes taste and judgment, not which application you're fluent in
Org charts will lag : teams structured around the design/build split will feel friction before they restructure
Rich Holmes, Department of Product
Why I picked this: It's the human counterpart to the model-layer commoditisation story - as the tools converge, the advantage shifts from tool fluency to judgment, the same theme running through the whole issue.
Execution: How Dropbox Built Its Internal Coding Agent, Nova
A concrete, build-it-yourself case study: how Dropbox designed, scoped, and rolled out an internal coding agent that engineers actually use. Useful if you're being asked "should we build our own?" and need to know what that really costs. Read the full article here.
💡 "The hard part wasn't the model - it was earning the engineers' trust one merged PR at a time."
| | Key Takeaways
Start narrow : Nova earned adoption by nailing a few high-frequency tasks before expanding scope
Trust is incremental : developer buy-in came from visible wins, not top-down mandates
Guardrails over autonomy : the agent's value came from operating safely inside existing review flows, not bypassing them
Internal tooling is a product : Dropbox treated Nova with the same discovery and iteration rigour as a customer feature
Dropbox Engineering
Why I picked this: It's the practical answer to the week's agent-outcomes theme - a real org showing what it takes to get from "agents are impressive" to "engineers rely on this daily".
🎙 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 convergence this week is uncomfortable: the AI frontier is further ahead than almost any organisation outside tech realises, the financial numbers are genuine, and most teams are making AI decisions based on a reality that no longer exists. Benedict Evans, Dax Raad, and the a16z team each approach this differently - but land in the same place.
🏗️ We're in 1997, not the industrial revolution - Benedict Evans's "controversial opinion" is that AI is only as big as the internet. He means it as a floor, not a ceiling - and the 1997 analogy is more useful for product planning than any other current framing.
📈 The exit threshold has 10x'd in 24 months - Top 1% exits went from $10B to $32B between February and May 2026 alone; a16z's David George expects $100B thresholds by September if Anthropic and OpenAI IPO as expected.
🔄 The productivity gap is widening - Dax Raad gives the honest version: ~20% improvement that feels like none. Ryan Lopopolo at OpenAI runs the equivalent of 5,000 engineers. Both are true - they're just not in the same company.
🤖 Background agents crossed a capability threshold - Devin's share of commits on its own repos went from 16% in January to 80% in March 2026. The async coding era is already here for the companies building within it.
🎯 AI made distribution the new moat - When anyone can build, the companies who sell and market win. Corgi's Nico says it explicitly; SaaStr data backs it up.
⚡ Inference margins are real, but the revenue accounting is circular - Dax Raad estimates 80 - 90% inference margins. Alex Kantrowitz documents how Google's Q1 profits were nearly half paper markups on Anthropic shares. These facts are related.
🧠 Enterprise AI is still on the developer's desktop - The shift to server-side agents via MCP gateways is the transition most enterprises haven't made, and it matters more than model choice.
🚀 The extreme culture thesis is back, with a $2.6B data point - Corgi Insurance hit $2.6B in two years with a founder who sleeps in the office and a team working seven days a week. Scalability remains open.
🌊 The net positive question is harder to sidestep - Teresa Torres's admission that the industry has "made the world worse in many ways" lands differently now that AI compounds the same structural choices.
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 🍽️.

