Sponsored by

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? 🥘

This week OpenAI decided chips are the new models and unveiled Jalapeno with Broadcom, while Anthropic shipped Tag, Science credits, and an MCP expansion - then brought Fable and Mythos 5 back from the export-control dead. Meanwhile Tesla and Sunrun casually wired together 16 GW of distributed solar (roughly 16 nuclear reactors' worth) to feed the data centres all this apparently needs, and AWS hired 11,000 juniors while the rest of the industry debates whether juniors are even useful any more. Maybe everything will be OK after all?

Your Menu

📰 Not Boring -> AI infrastructure goes vertical, Claude ships a triple, developer craft evolves around agent loops, and AWS makes a contrarian talent bet
⌚️ Productivity Tapas -> HN Trends for developer sentiment, GolemUI for portable forms, VC Heatmap for cold outreach, and a 13-year-old's ant tracker
🍔 Blog Bites -> Why AI adoption is a trust problem not a tooling one (Lenny's), Return on Tokens as a PM metric (McCormick), and the design tells that give away AI-built interfaces (Growth.Design)
🎙️ Pod Shots -> Eight conversations, one heresy: use less AI, protect the judgment the machine cannot give you

Let's go 🚀

📰 Not boring

🤖 AI's Build Stack Gets Real

The chip race is no longer about who has the best model - it is about who owns the power, the silicon, and the inference layer underneath it. OpenAI building custom chips while Tesla wires 16 GW of distributed solar into data centres tells you where the real bottleneck moved.

🏷️ Claude's Week - Tags, Science, and the Agent Layer

Anthropic shipped more surface area in one week than most companies ship in a quarter. Tag redesigns the conversation model, Claude Science buys goodwill in academia, and the MCP ecosystem keeps expanding. The pattern: stop selling the model, start selling the workflow around it.

🛠️ Developer Craft - Loops, Parsers, and PR-Opening Agents

The developer tooling conversation has moved past "can AI write code" to "what does the development workflow look like when it can." Loops replace prompts. PRs replace chat. Token costs become a design constraint. The craft is not dying - it is being rebuilt around different primitives.

🏭 Product Craft in the Agent Era

LinkedIn fixing 200 bugs with cognitive-model QA agents, Figma shipping code layers, and a new job title - "vibe architect" - entering the vocabulary. The product craft is not shrinking. The tools changed, and the roles are following.

💰 Money, Power, and the Talent War

AWS hiring 11,000 juniors while the rest of the industry debates whether juniors are obsolete is the kind of contrarian bet worth watching. Either Amazon sees something everyone else is missing, or they are building a very expensive insurance policy against AI not replacing entry-level work.

🔬 Frontier Science Meets Product

A brain-imaging startup built by non-scientists who worked backwards from "we want brain interfaces" and taught themselves ultrasound physics. A microreactor running at full power for 24 straight hours. The frontier is moving because outsiders keep refusing to respect the boundaries of what they are supposed to be qualified for.

📱 Platform Moves and Creator Tools

Agentic commerce APIs opening to all developers is the sleeper move here. When agents can buy things autonomously, the entire conversion funnel that PMs have spent a decade optimising becomes a machine-to-machine negotiation. The interface shifts from "convince a human to click" to "convince an agent your API is reliable."

How Much Is Your Billing Lag Actually Costing You?

Most SaaS finance teams know their billing process isn't perfect. Few know what it's actually costing them.

Answer 5 quick questions — contracts signed per month, ACV, days to first invoice, error rate, DSO — and the Tabs Billing Lag Calculator gives you a dollar figure benchmarked against top SaaS companies.

It takes two minutes. The number might surprise you.

Calculate your billing lag and see where you stand.

Calculate your billing lag

⌚ Productivity Tapas: Time-Saving Tools & Workflow Automation

  • Hacker News Trends: Google Trends for Hacker News - charts how often any topic, tool, or person has surfaced across 18 years of HN comments and posts. Useful for tracking developer sentiment and spotting emerging tools before they hit mainstream. Free.

  • GolemUI: Declarative form engine that renders dynamic forms from JSON across React, Angular, Vue, and Lit. Define once, deploy anywhere - forms are portable, versionable, and diffable. Just hit v1.0. Free and open-source.

  • LingoChunk: Converts native-language audio into Anki flashcards with native-speaker audio snippets. Upload a podcast episode in your target language and get study-ready flashcard decks back. Practical for PMs working across language markets.

  • VC Heatmap: Interactive heatmap of 3,400+ VCs who are open to cold emails, filterable by stage, sector, and geography. Built on Apparent's founder-investor matching platform. Free to browse.

  • Formicarium: An ant colony tracker built by a 13-year-old. Included because it is a genuinely charming example of a kid shipping a real product - and because tracking complex systems with simple tools is basically what PMs do every day.

    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

Leadership: Your AI Strategy Has a Trust Problem, Not a Tooling Problem

The argument here cuts against the grain of most AI rollout advice: teams aren't stalling on AI adoption because they lack the right tools - they're stalling because they don't trust what the tools mean for how their work is judged. Read the full article here.

💡 The real blocker to AI adoption is rarely the model - it is whether people trust how their judgment will be measured once the machine is in the loop.

| | Key Takeaways

  • Trust, not tooling : Rolling out another AI tool does nothing if the team suspects it exists to measure or replace them. Adoption follows psychological safety, not feature count.

  • Name the fear : People quietly worry AI makes their contribution invisible. Leaders who address that directly get faster, more honest adoption than those who just mandate usage.

  • Measure outcomes, not usage : Tracking who uses the tool most rewards theatre. Tracking what the tool helped ship rewards the behaviour you actually want.

  • Model the behaviour : Teams take their cue from leaders who use AI to augment their own judgment openly, rather than hiding it or wielding it as a stick.

Lenny's Newsletter

Why I picked this: Every team wrestling with AI adoption reaches for a new tool when the real gap is trust. It connects directly to the "use less AI, use it more deliberately" thread running through this week's Not Boring section - the constraint was never the tooling.

Data & AI: Return on Tokens - AI Is a Compiler, Not a Runtime

Packy McCormick reframes the AI cost conversation with a concept that should change how you budget: Return on Tokens. Stop thinking of AI as a service you run and start thinking of it as a compiler that produces an asset. Read the full article here.

💡 "The question isn't how many tokens you spent. It's what durable asset those tokens compiled into existence."

| | Key Takeaways

  • Compiler vs runtime : When AI generates a codebase, a dataset, or a strategy document, the tokens are a one-time compilation cost. The output persists and compounds. Treat the spend like capex, not opex.

  • ROT as a metric : Return on Tokens forces you to ask what lasting value each batch of spend produced, rather than optimising for cost-per-token in isolation.

  • The subsidy window is closing : Current token pricing is artificially low. Teams building habits around cheap frontier inference are accumulating a dependency they will pay for later.

  • Defensibility lives in the compiled artefact : The model is rented. The evals, the fine-tuned dataset, and the workflow built on top of it are yours. Invest in the layer you keep.

Packy McCormick, Not Boring

Why I picked this: This pairs perfectly with Mike Krieger's "tokens are not outcomes" line from Pod Shots #142 this week. McCormick gives PMs a concrete mental model - compiler, not runtime - for making the case to leadership that AI spend should be evaluated like an investment, not a subscription.

Design & UX: The Tiny Tells That Give Away AI-Built Design

Growth.Design dissects the subtle patterns that make AI-generated interfaces feel uncanny - the spacing that is too perfect, the copy that is too smooth, the interactions that follow templates instead of user intent. Read the full article here.

💡 "The AI knows the rules of design. It just doesn't know the reasons behind them."

| | Key Takeaways

  • Uncanny uniformity : AI-generated layouts tend toward mathematical perfection - equal spacing, symmetrical grids, balanced type hierarchies. Real interfaces have intentional imperfections that guide the eye.

  • Template-shaped copy : AI writes headlines that follow patterns it has seen. The result reads like every SaaS landing page merged into one. Human copy has rough edges that signal authenticity.

  • Missing micro-decisions : Experienced designers make hundreds of small judgment calls - when to break a grid, where to add friction, which element deserves visual weight. AI follows the median.

  • The vibe architect gap : This is exactly why the "vibe architect" role is emerging. Someone has to provide the creative direction that turns competent AI output into something with a point of view.

Growth.Design

Why I picked this: The "vibe architects" story from this week's Not Boring section describes the job. This article shows you the specific failure mode it solves - AI design that is technically correct but emotionally flat. If you are shipping AI-generated interfaces, this is the quality bar to check against.

🎙 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

🤖 Pod Shots #142: Use Less AI

Everyone's chasing a smarter model. The teams actually pulling ahead are winning somewhere far less glamorous: utilisation, craft, cheap experiments and the discipline to copy what works. Across seven recent conversations - from the people behind Anthropic's compute strategy, Microsoft, Zynga, Clay, the iPhone and xAI - one uncomfortable pattern keeps surfacing: the bottleneck is almost never the capability. It's the waste sitting right next to it.

Across eight conversations - Anthropic's own Claude Code lead, the head of Anthropic Labs, Meta, a16z, Supabase, two AI founders and the Mind the Product debrief - the same heresy kept surfacing. The bottleneck was never the building. And the teams pulling ahead are the ones who noticed that maximising output and maximising value are not the same thing.

In this roundup I’ve pulled the sharpest idea from each show and woven them around that spine, so you get the whole argument in eight minutes instead of eleven hours. Where the guests agree, we have stacked them. Where they genuinely disagree, we have left the tension standing.

💡 Top tip - read the TL;DR, then jump to the section closest to your current problem. "Use less AI" and "tokens are not outcomes" are the load-bearing ones.

Featured in this round-up:

  • Fiona Fung, Anthropic (Manager, Claude Code & Cowork teams), "What happens after coding is solved?", Lenny's Podcast - 🎧 Listen - 📆 22-06-2026

  • Kevin Weil, formerly CPO at OpenAI, "AI is crossing the frontier of human knowledge", a16z Podcast - 🎧 Listen - 📆 26-06-2026

  • Paul, CEO of FOMO, "$94M, why 1-1s are BS, and the death of moats", 20VC - 🎧 Listen - 📆 27-06-2026

  • Jagjeet Chawla, Meta (Feed, Reels & Search), "How Meta is reinventing product management", The Skip - 🎧 Listen - 📆 26-06-2026

  • Yash Patil, founder of Applied Compute, "Own or be owned: why every company needs its own AI model", The Generalist - 🎧 Listen - 📆 24-06-2026

  • Paul Copplestone, CEO of Supabase, "How Supabase became essential AI infrastructure", First Round Review - 🎧 Listen - 📆 26-06-2026

  • Mike Krieger, head of Anthropic Labs (co-founder, Instagram), "Building AI-native products", Big Technology Podcast - 🎧 Listen - 📆 25-06-2026

  • Emily Tate, CPO and MD of Mind the Product, "A deep dive into the state of product in 2026", Mind the Product - 🎥 Watch - 📆 25-06-2026

🕒 Estimated reading time: 8 mins. Time saved: 11+ hours! 🔥

Not your topic this week? Try these five instead:

Key insights from the full round-up:

  • 🛑 "Use less AI" - FOMO's CEO, building a whole product in three weeks with Claude Code, still tells computer-science students to use less of it: the best engineers learned the craft without it, and he avoids it for his own writing so he does not lose the skill.

  • ⏳ Going faster can mean going slower - Yash Patil's warning: ship on a foundation you did not think through and "by going faster you can actually be going a lot slower." The bill arrives later.

  • 💸 Tokens are not outcomes - Mike Krieger's reframe: compare your top-ten token users against your top-ten most productive people. Anthropic weighs intelligence, effort and token efficiency, not raw spend.

  • 🎓 Nobody can yet teach the next generation - Fung's open worry: how do you give a junior the "double-click" understanding that used to come from years of typing code, when the agent does the typing?

  • 🧱 The moat fight has three camps now - own your own model (Patil), own a non-AI asset like a social graph or brand (FOMO, Tate), or own the operating layer and refuse lock-in (Copplestone). They cannot all be right.

  • 🤖 "AI-first" is becoming a liability - Emily Tate, outside the tech bubble: position yourself as AI-first and you trigger real backlash. "My mum doesn't want to figure out Claude Code."

  • 🔬 The capability is still racing ahead - Kevin Weil: models solved ten to twelve open maths problems in January, going beyond the frontier of human knowledge. The backlash is about how we use it, not whether it is real.

  • 🫂 Everyone is more productive and more alone - the quiet counter-melody: agentic work is lonely, remote work is draining the joy, and the labs are now engineering human reconnection back in.

EXPLORE THE FULL GUIDE →

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 🍽️.

Reply

Avatar

or to participate

Keep Reading

View more
caret-right
© 2026 Product Tapas..
beehiivPowered by beehiiv