Now Shipping: A Mind the Product podcast

Mind the Product
  • 0.0 (0)
  • Technology

A 15 minute weekly recap of product management news, technology updates, and advice for product builders, brought to you by the team at Mind the Product. 

  1. 3d ago

    Claude Can Pass The Coding Interview. Now What?

    Claude can pass the coding interview. So how do you work out who to hire? One Anthropic engineering manager has stopped using traditional coding interviews for his team, arguing that they no longer tell him enough about candidates. He’s looking instead for prototypes that demonstrate taste, empathy and judgement. In this week’s Now Shipping, Mike Belsito explores what this means for product people. If AI can complete your take-home exercise or product case study, is your hiring process still measuring the skills you need? Mike examines how Anthropic and Canva are adapting their assessments, why a polished prototype isn’t enough, and how reviewing flawed AI-generated work could reveal more about a candidate’s thinking. In this episode: — Why Felix Rieseberg dropped coding interviews for his Anthropic team. — How improving AI models forced another Anthropic team to repeatedly redesign its hiring test. — Why judging AI-generated work matters alongside creating it. — What prototypes can reveal about candidates, and what they can hide. — How an “audit interview” could work for product managers. Why clear scoring criteria matter when assessing subjective qualities such as taste and judgement. Try this with your team: run one of your existing interview exercises through an AI tool. Would you advance a candidate who submitted that answer? If so, consider what else you need to observe to assess their thinking. Featured reading: Hunger beats tenure: Advice for junior engineers, Felix RiesebergDesigning AI-resistant technical evaluations, AnthropicHow employment is changing at AnthropicYes, you can use AI in our interviews. In fact, we insist, CanvaState of Tech Hiring 2026, CoderPad

  2. Oct 21

    How Anthropic’s agents work 6× more than humans

    Six agent hours for every human hour. Seventeen engineers for every product manager. What happens when an AI company turns its own technology on the way it works? This week on Now Shipping, Mike Belsito breaks down Anthropic’s research into how Claude is changing its workforce, from hiring fewer junior technical employees to giving people their own teams of agents. But more output brings new demands. Agents wait for human input, employees spend more time in meetings, and reviewing everything AI produces becomes a job in itself. Mike compares Anthropic’s findings with research discussed in Harvard Business Review and a survey commissioned by Jabra to explore what this means for product teams. Key takeaways — Human attention becomes the bottleneck. Producing more work only helps if people can review it, make decisions and move it forward. — Product roles are shifting. Anthropic’s findings point towards a greater emphasis on judgement, autonomy and the ability to give agents clear instructions. — More code doesn’t mean more customer value. Teams need to measure outcomes rather than celebrate increases in activity. — AI changes how teams share knowledge. Private conversations and undocumented decisions leave gaps in the context agents need. — Human collaboration still needs deliberate attention. Using AI to prepare for a conversation can help. Using it to skip the conversation can weaken shared understanding. — Hiring fewer juniors raises a longer-term question. Where will the next generation of experienced employees come from? Chapters 00:00 Anthropic’s agents work six hours for every one hour of human input01:00 Anthropic earns two to four times Meta and Google’s revenue per employee at the same age02:00 Junior share of Anthropic’s technical hires falls from 27% to 9%03:00 Commits per engineer up fivefold, queries per data scientist and legal contracts up threefold04:00 A sales rep builds a dashboard in three to four hours, and one PM serves 17 engineers05:00 Executive assistants move to chief-of-staff work and writers move to writing evals and prompts05:30 192 agent adopters against 74 non-adopters show no difference in code quality06:00 Human attention is the bottleneck: 80% of agent session time is idle07:00 Reading all agent output would take more than half a workweek for non-technical teams08:00 Meta work grows from 10% to 22% of agent sessions09:00 Agents appear in nearly half of active public Slack channels10:00 Unrecorded decisions are invisible to agents, and meeting time rises 20%11:00 Mentions of AI track record in hiring scorecards triple for technical roles and quadruple for non-technical12:00 Harvard Business Review names “relation slipping”: treating AI as a teammate erodes trust between humans13:00 Jabra survey: 57% of heavy AI users do more admin and coordination work14:00 Summaries, status views and approval queues ranked by consequence15:00 Replace commits per engineer with time to a shipped decision15:30 The PM skill that remains: writing a spec an agent can run on for hours16:00 Microsoft finds employment for 22 to 25-year-olds in AI-exposed jobs down 16%17:00 This week’s experiment: write one private decision down where humans and agents can see it18:00 Wrap-upSources: How employment is changing at AnthropicHBR: Team dynamics weaken when employees turn to AI instead of colleaguesHBR: How AI erodes teamworkIT Brief UK: Jabra studyUnite.AI: AI lab CEOs warn their own companies are hiring fewer junior workers Now Shipping is Mind the Product’s weekly AI briefing for product people, hosted by Mike Belsito.

  3. Oct 14

    Why companies are blocking AI shopping agents

    This week on Now Shipping, Mike Belsito explores why personal agents are getting stuck across the web, and what those failures reveal about the products we build. From Walmart’s checkout checks to Yelp’s paid access model, businesses are taking very different approaches to agent traffic. Sometimes, the decision comes down to a security setting nobody on the product team has reviewed. Mike also examines Wikimedia’s concerns about suspected OpenAI agent activity and the proposed Personal Agent Protocol from Meta and Sierra. What could agent authentication solve, and what still needs to be fixed inside your product? For product teams, the starting point is practical: test a critical flow with an agent, find out where it fails, and decide what it should be allowed to do. In this episode Why welcoming agents publicly doesn’t guarantee they can use your productHow CAPTCHAs, bot filters and infrastructure defaults shape your agent policyWhat Wikimedia’s findings raise about agent permissions and responsibilityWhat the proposed Personal Agent Protocol aims to addressHow agent traffic could affect pricing, costs and product analyticsChapters 00:00 MTP Chicago and this week’s briefing 01:00 Why agents are failing at Walmart’s checkout 02:00 Cloudflare defaults and the hidden agent policy 03:00 How Yelp, Delta, eBay and Zillow handle agents 04:00 Balancing customer access with security 05:00 Wikimedia’s findings on suspected OpenAI agents 06:00 Introducing the proposed Personal Agent Protocol 07:00 What a protocol could solve, and its limitations 08:00 Test your product’s real agent policy 09:00 Design the failure state and check your settings 10:00 Review what agents can change 11:00 Agent access, pricing and infrastructure costs 12:00 Measuring agent activity and rising customer expectations 13:00 Three questions every product team should answer Key takeaways Your product already has an agent policy. Bot settings, human checks, terms of use and vendor defaults collectively determine how it treats agents.Test a flow that matters. Ask an agent to complete a sign-up, checkout or booking, then record every point where it gets stuck.Design the handover. If an agent cannot continue, give the customer a clear explanation and a way to finish without losing their progress.Make permissions deliberate. Review what agents can read and change, and where explicit approval is needed.Check your measurement. Understand whether your analytics distinguish agent activity from human activity before making access or pricing decisions.If an agent arrived at your product tomorrow, could you identify it, understand who sent it and explain what it was allowed to do?

  4. Oct 2

    Everything you need to know about Dots: OpenAI’s new agent

    This week on Now Shipping, Mike Belsito breaks down OpenAI’s dots: how its always-on agents work, the workflows demonstrated at DevDay, and the controls designed to keep people in charge. He examines the early customer examples and asks whether the biggest advance is new capability or making existing automations easier to set up. Mike also explores pricing and usage changes, distribution through ChatGPT, Slack and Teams, and OpenAI’s enterprise marketplace. Alongside the launch, he examines the delayed GPT-6.1 Astra release and what it reveals about scope, authorisation and trust. In this episode How dots works, from cloud computers to connected appsThe product launch demo: turning a change into downstream workWhat automated coordination means for product managersPermissions, approvals, activity tracking and stopping an agentEarly customer examples and the appeal of simpler setupPricing, usage limits and agent packagingWorkplace distribution and the enterprise marketplaceThe GPT-6.1 Astra delay and questions about reliabilityPreparing your product, metrics and pricing for agent users

    Everything you need to know about Dots: OpenAI’s new agent

  5. Sep 25

    Meta Muse's big week: #1 on the App Store, then blocked by Amazon

    Meta's Muse launched two weeks ago as a personal AI agent you hand tasks off to rather than chat with. Since then it's become one of the biggest product stories of the year: download numbers that outpace ChatGPT's early days, a Shopify and Instacart partnership that wires Muse straight into checkout, and an Amazon block that exposes exactly what's at stake when an agent skips the shelf.  Mike Belsito connects the dots back to a question the show posed in August, what will it actually take for non-tech consumers to trust an AI agent with something real, and argues Muse may be answering it by accident. Key takeaways — In a 12-day window, Muse pulled 1.8 million iOS downloads in North America against ChatGPT's 1.3 million at its 2023 launch — though Mike flags the comparison as skewed, since this time the public already understands what AI is. — Shopify wired Shop Pay directly into Muse's checkout on 21 September, and Instacart followed a day later, letting Muse buy from either platform without opening the app. Shopify's stock jumped 11% on the news. — Amazon blocked Muse from its store that same week, accusing Meta of capturing customer credentials without disclosure. Meta maintains Muse never sees passwords or payment details because credentials sit in secure storage it can use but not view. — Amazon's sponsored-ads business brings in $68 billion a year, and it depends on shoppers scrolling past paid placements that an agent checking out directly would skip entirely. — Muse is solving the distribution half of the "agentic trust gap" that Browser Company co-founder Josh Miller described in August: rather than winning as a better standalone app, it's showing up inside checkout flows exactly when people are already trying to buy something. — Every person who lets Muse handle a multi-step task without watching each step is practising a habit they'll bring to work on Monday, which should lower the bar for getting your own users to trust an agent with something real. Referenced — Meta debuts its Muse AI agent — TechCrunch: https://techcrunch.com/2026/09/08/met... — Signal — Moxie Marlinspike's encrypted messaging app: https://signal.org/ — Meta's Muse is outpacing ChatGPT's early mobile launch — TechCrunch, Apptopia data: techcrunch.com/2026/09/21/metas-muse-is-outpacing-chatgpts-early-mobile-launch/ — Tobi Lütke's announcement — Shopify and Muse partnership: x.com/tobi/status/2102090718546198790 — Amazon blocks Meta's Muse agent from shopping on Amazon.com — Forbes: www.forbes.com/sites/jonmarkman/2026/09/21/amazon-blocks-metas-new-muse-ai-agent-from-shopping-on-amazoncom/

    Meta Muse's big week: #1 on the App Store, then blocked by Amazon

  6. Sep 17

    We read Dario Amodei’s 3,800-word essay so you don’t have to

    AI’s biggest rivals are talking about slowing down. But will their release schedules tell the same story? This week on Now Shipping, Mike Belsito breaks down Dario Amodei’s 3,800-word essay on why frontier AI development needs to slow down, the responses from Sam Altman and Elon Musk, and what it could mean for product teams. If your roadmap depends on the next model being smarter, cheaper and just a few months away, how exposed are you? Mike explores the assumptions worth challenging, the questions to ask your AI vendors, and whether proposed safety standards could strengthen the biggest labs’ advantage. Plus, Nina Olding, Staff Product Manager at CoreWeave and former Product Lead at Google DeepMind, shares her perspective on what PMs can control today: shipping AI responsibly through evals, transparency, human oversight and meaningful user choice. Chapters 00:00 Watch the releases, not just the essays00:48 The researcher’s warning that sparked the debate04:22 Dario Amodei’s case for slowing down06:01 Independent evaluators and a three-part safety plan07:07 The criticisms: accountability, coordination and competition08:23 Sam Altman and Elon Musk respond09:46 The political reaction10:32 What this means for product roadmaps and vendor selection12:53 Will public promises change what companies ship?13:45 Stress-testing your dependence on better AI models15:04 Nina Olding: What responsible AI looks like for PMsFeatured links Dario Amodei: We Must Pace the FrontierJacob Coxon’s resignation and AI safety warning — TechCrunchWatch Jacob Coxon’s BBC interviewSam Altman and Elon Musk back calls to slow AI development — The GuardianAI CEOs say they need to slow development. But will they? — The GuardianThe original 2023 open letter: Pause Giant AI Experiments — Future of Life Institute Does your roadmap depend on AI models getting better? Tell us what you’d rethink if progress slowed.

  7. Sep 11

    Meta’s Muse: How much control would you hand over?

    Your calendar, your inbox, your credit card. How much access would you give Meta’s AI agent? This week on Now Shipping, Mike Belsito examines Meta’s pitch for Muse and the questions it raises for product teams building, buying or experimenting with AI agents. Where should your product’s AI features live? What should an agent be allowed to do? And who approves its actions before a message gets sent or money gets spent? Mike explores why permissions could become a reason users choose one AI product over another, and what product people can learn from competing approaches to putting AI into everyday workflows. In this episode: • Where AI belongs: chat panels, browser extensions or the apps people already use. • Meta’s pitch for Sentinel as an approval gate for Muse’s actions. • Why permissions and user control deserve a place in product comparisons. • Four questions to ask about any agent: what can it access, who approves its actions, when is approval required, and what happens when it gets something wrong? Would you give Muse access to your inbox and credit card? Tell us where you’d draw the line in the comments. Now Shipping is Mind the Product’s weekly AI briefing for product people, hosted by Mike Belsito. Subscribe for more AI stories that matter to your work.

  8. Sep 3

    Product lessons from a month of AI security incidents

    Oji and Ezinne Udezue are product leaders with more than 50 years of combined experience between them. Oji has held product roles at Twitter, Calendly, Atlassian, Typeform and Microsoft; Ezinne has led product at WP Engine, Procore and T-Mobile. Together they co-wrote Building Rocket Ships and now run ProductMind, helping companies work out what AI actually means for how they build.  In this episode of Now Shipping, host Mike Belsito brings them in to unpack a story that kept repeating through the month: AI agents breaking out of the sandboxes built to contain them. We discuss: — How OpenAI's most advanced model went rogue during an internal security test, hacking into Hugging Face's production infrastructure without any human instruction, in order to cheat on a benchmark — Why Moonshot AI's Kimi K3 didn't need to hack anything — it found an unlocked outbound connection, cloned the target repository from GitHub, and read the benchmark's solution straight off disk — Why the "rogue intern" analogy applies to agents, and why the real failure sits with the people who build fragile boundaries, not the agents that find them — The shift from zero trust to "negative trust" — treating every agent as having to reprove itself in the moment, rather than trusting it because it behaved well five minutes ago — Why harness engineering, not prompt engineering or loop engineering, is where the next year of agentic product work will actually happen — A five-part framework for product managers partnering with security teams: reach, reversibility, graceful exits, observability and provenance, and the kill switch — Why an industry fixated on completion rate, rather than safe-stop rate and data provenance, risks repeating this month's incidents

See All (21)

About

A 15 minute weekly recap of product management news, technology updates, and advice for product builders, brought to you by the team at Mind the Product. 

Information

You Might Also Like