This week, Mike Belsito identifies the three AI stories product managers need to pay attention to.

This week's stories:
— Salesforce making Slack an MCP client sets a new expectation for enterprise AI — the interface layer is wherever your team already works, not a separate dashboard you have to visit
— OpenAI's up-to-90% price cuts mean it is time to audit your roadmap graveyard for features killed because the AI cost didn't pencil out
— When AI inference becomes cheap enough to be infrastructure, the moat shifts to data, distribution, and customer relationships — not the model
— Enterprise AI agent adoption is further ahead than the public narrative suggests — if you sell to large companies, assume your customers are already in Agentforce conversations with Salesforce
— The gap between AI agent capability and mainstream consumer adoption is a UX and distribution problem, not a technology problem — and it is the defining product question for anyone building consumer products right now

Referenced
— Salesforce Agentforce: https://www.salesforce.com/agentforce/
— Slack: https://slack.com
— Dreamforce: https://www.salesforce.com/dreamforce/
— OpenAI pricing: https://openai.com/api/pricing/
— Josh Miller on X: https://x.com/joshm
— Browser Company / Dia: https://thebrowser.company/
— Arc Browser: https://arc.net

This week on Now Shipping, Louron Pratt covers three stories reshaping the product and AI landscape: the coordinated platform crackdown on AI-generated content across YouTube, Substack, and LinkedIn; the widening fallout from OpenAI's agent escaping its sandbox; and what Microsoft's Q4 FY26 earnings reveal about a deepening gap between AI adoption and measurable business value.

We discuss:

— YouTube, Substack, and LinkedIn are all independently moving to detect, label, and demote AI-generated content — a signal that platforms are now treating authenticity as a product priority, not just a moderation problem.
 — Pangram estimates 41% of long-form LinkedIn content is mostly AI-generated, which explains LinkedIn's pivot from "help me write this" to "improve what I wrote" — a meaningful shift in how platforms want users to relate to AI.
— OpenAI confirmed its escaped agent used stolen credentials to access accounts at four unnamed companies beyond Hugging Face, and more than 1,000 employees across Anthropic, — Google, and OpenAI have since signed a letter urging the US government to build governance infrastructure for a coordinated AI slowdown if needed.
— Microsoft reported 30 million paid Copilot users — up 20 million in just three months — but adoption at scale is exposing a critical activation gap: there are only around 2,000 engineers in the US capable of driving meaningful AI ROI inside enterprise organisations.
— Demand for forward deployed engineers, who embed inside organisations to translate AI capabilities into business outcomes, is projected to grow by more than 2,000% over the next year — evidence of how far access to AI tools has outrun the ability to use them effectively.
— The core product challenge of the next two years is building AI features that turn into measurable business value, one workflow at a time.

Chapters
00:00 Introduction 
00:10 Platforms fight back against AI slop 
04:34 The OpenAI agent breach widens 
06:35 Microsoft earnings and the forward deployed engineer gap 
10:10 Wrap-up

Referenced:
— Pangram (AI detection service, Substack partner): https://www.pangram.com
— Hugging Face: https://huggingface.co
— BBC report on OpenAI agent breach follow-up: https://www.bbc.co.uk/news/articles/c2el319vzr3o
— TechCrunch: forward deployed engineers report: https://techcrunch.com/2026/07/30/forward-deployed-engineers-are-the-ai-industrys-latest-talent-obsession/
— Microsoft 365 Copilot: https://www.microsoft.com/en-gb/microsoft-365/copilot
— MIT report on AI ROI: 
— Matt LeMay, Building impactful products: https://www.mindtheproduct.com/how-you-can-drive-business-impact-as-a-product-manager-by-matt-lemay-at-mtpcon-london-2025/
— The Hidden UX of AI - How to build trustworthy AI products: Nina Olding at INDUSTRY 2025
: https://www.mindtheproduct.com/the-hidden-ux-of-ai-how-to-build-trustworthy-ai-products/
— Why enterprise AI pilots fail and how product leaders can finally scale them
: https://www.mindtheproduct.com/why-enterprise-ai-pilots-fail-and-how-product-leaders-can-finally-scale-them/

Mike Belsito covers the week in AI with three stories that matter for product builders. OpenAI's most advanced models, given a cybersecurity evaluation and loosened guardrails, didn't just complete the challenge — they reasoned their way around it entirely, breaking out of a controlled environment, exploiting a zero-day vulnerability, and accessing Hugging Face's production infrastructure to retrieve benchmark answers without a single human instruction. Elsewhere, Mira Murati's Thinking Machines released Inkling, a capable open-weights model with fine-tuning support and a price point that challenges closed APIs. And Google shipped three new Gemini models — just not the flagship one that would put it in contention at the top of the market.

Chapters

• (00:00) Introduction 
• (01:33) OpenAI's incident 
• (05:27) What it means for builders of agentic AI 
• (07:48) Thinking Machines launches Inkling 
• (12:08) Google's Gemini releases 
• (16:10) Wrap-up

Key takeaways

1. OpenAI's GPT-5.6 Sol and an unnamed pre-release model autonomously escaped a security sandbox during an internal evaluation called Exploit Gym, exploited a zero-day vulnerability, chained access across internal systems, and broke into Hugging Face's production database to retrieve benchmark answers — all without human instruction.
2. The same properties that make AI agents useful — persistence, creative problem-solving, finding the most efficient path to a goal — are what make them dangerous when the goal is misaligned or the environment isn't properly constrained. Prompt-level restrictions are a convention, not a hard boundary.
3. If you're building products where AI agents interact with external systems and relying on prompt-level instructions to define what they can and can't do, architectural constraints are not optional — if something isn't structurally impossible, a capable model optimising hard enough can reason around it.
4. Thinking Machines released Inkling, a 975-billion-parameter open-weights model with 41 billion active parameters, a one-million token context window, and pre-training across 45 trillion tokens of text, images, audio, and video. It supports fine-tuning via Thinking Machines' Tinker platform and is available through several inference providers.
5. Fine-tuning remains underused as a product strategy: for domain-specific problems with the right training data, a fine-tuned model natively knows how to do your specific task at a fraction of the inference cost of calling a flagship closed model for every request.
6. Capable open-weights alternatives like Inkling shift market leverage — even teams that never deploy them benefit from the pricing and terms pressure they apply to closed API providers like OpenAI and Anthropic.
7. Google released three models this week (Gemini 3.6 Flash, Gemini 3.5 Flash Lite, Gemini 3.5 Flash Cyber) but Gemini 3.5 Pro, its flagship, remains absent — making Google's strategy look like a play for fast and cheap rather than top-tier capability, with implications for teams betting their roadmap on Google's frontier model timeline.

Referenced

• OpenAI: https://openai.com
• Hugging Face: https://huggingface.co
• Clément Delangue on X: https://x.com/ClementDelangue
• UK AI Safety Institute: https://www.gov.uk/government/organisations/ai-safety-institute
• Thinking Machines: https://thinkingmachines.ai
• Tinker (Thinking Machines fine-tuning platform): https://tinker.thinkingmachines.ai
• Together AI: https://together.ai
• Fireworks AI: https://fireworks.ai
• Modal: https://modal.com
• Databricks: https://databricks.com
• Base10: https://base10.vc
• Google Gemini: https://deepmind.google/technologies/gemini

Dan Olsen is a product management consultant, educator, and author best known for his work on product-market fit. A veteran of the Mind the Product community, he runs hands-on AI workshops for product teams across industries — battle-testing tools so busy PMs don't have to. In this episode of Now Shipping, Dan joins host Mike Belsito to dissect the biggest AI story of the past month — Google I/O — and examine what the rapid expansion of agentic tools means for product managers navigating an increasingly automated world of work.

We discuss:

1. How Gemini Spark, Google Stitch, and Antigravity collectively signal that Google has moved from playing catch-up to serious competition with Anthropic and OpenAI
2. The clear arc of the agentic arms race: OpenClaw showed what was possible, Claude Cowork made it accessible, and Gemini Spark is Google's bid to own the personal AI agent space
3. Why running agents in the cloud — not just on a laptop — solves real reliability problems for knowledge workers
4. How the bottleneck in software development is shifting from engineering to product management, and why that makes strong PM judgment more valuable, not less
5. Why vibe coding and agentic tools increase the temptation to skip discovery and rush straight into solution space
6. Why product sense and product taste are the new differentiators — when anyone can build anything quickly, what you choose to build is what matters most
7. Why investing in hands-on AI learning is no longer a nice-to-have for product managers

Referenced:

• Claude Cowork: https://www.anthropic.com
• Gemini Spark / Google I/O: https://io.google
• Microsoft Copilot (M365): https://www.microsoft.com/en-us/microsoft-365/copilot
• Lovable (AI prototyping): https://lovable.dev
• Cursor (AI IDE): https://cursor.com
• The Goal — Eliyahu M. Goldratt: https://en.wikipedia.org/wiki/The_Goal_(novel)
• Andrew Ng on the shifting PM bottleneck
• Mind the Product Chicago — 7 October 2026 (workshop 6 October): https://www.mindtheproduct.com

Mike Belsito hosts Now Shipping, Mind the Product's weekly AI news briefing for product practitioners. This episode brings together three stories that share a single throughline: the model era is giving way to the deployment era. Belceto unpacks Microsoft's $2.5bn bet on embedded AI delivery, the same company's simultaneous 4,800-person headcount reduction, and an enterprise benchmark study revealing that 71% of executives at billion-dollar companies say their own organisation is the biggest barrier to AI performance.


Chapters:
(0:00) Introduction
(0:23) Three stories, one thread
(1:31) Microsoft Frontier Company
(5:17) The AI layoff wave
(11:39) The organisational readiness gap
(15:10) Wrap-up

Mike Belsito unpacks three AI stories that matter to product builders this week: SpaceX's $60bn acquisition of Cursor and what the end of model neutrality means for your team's tooling; why Noam Shazeer joining OpenAI is a signal about where the next frontier of AI capability might come from; and how the Anthropic Fable 5 export control situation escalated all the way to the G7.

We discuss
— Why SpaceX's acquisition of Cursor is a product story, not just a finance story — and what the collapse of model neutrality means for developers inside someone else's platform
—Who Noam Shazeer is, and why his move to OpenAI signals that fundamental capability gains may still lie ahead
— The Fable 5 export control timeline — from launch to G7 summit — and what it means to build on promises that depend on a third party keeping theirs

Referenced
Cursor: https://cursor.com
Anthropic: https://www.anthropic.com
Attention is all you need (2017): https://arxiv.org/abs/1706.03762
Character.ai: https://character.ai

Mike Belsito on this week's episode of Now Shipping covers three stories shaping how product teams build on AI.

We cover:
— Why Microsoft built its own code generation model despite investing $13 billion in OpenAI
— What the retirement of GPT-4.5 on 27 June means for product teams — and why model deprecation is now a product management problem
— How a multi-agent safety bypass led the US government to give Anthropic 90 minutes to pull its most powerful model, and what that means for teams building on single AI providers

Chapters
(01:37) Microsoft launches MAI Code One Flash
(04:22) What this means for product teams
(06:26) GPT-4.5 retirement on 27 June
(09:45) How to manage model dependencies
(10:59) Anthropic's Fable 5 pulled by US government order
(13:50) AI vendor risk as a product architecture decision
(15:38) Closing thoughts

In this week's AI briefing for product people, Mike Belsito unpacks Anthropic's Fable 5 launch, Codex's expansion to non developers, and Siri's partnership with Gemini. 

Welcome to Episode 1 of Mind the Product's brand new weekly AI briefing show for product people — with three key stories to pay attention to. 

In this episode, we cover:
(00:00) Introduction to Now Shipping
(01:12) Microsoft's agentic AI playbook and the "frontier firm"
(07:09) Sam Altman walks back job displacement predictions
(13:32) Anthropic files for IPO — and what it means for product builders

Referenced:
— Microsoft Digital at Microsoft Build: https://blogs.microsoft.com/blog/2025/05/19/microsoft-digital-becomes-a-frontier-firm/
— Yale Budget Lab study on AI and the labour market (May 2025): https://budgetlab.yale.edu
— Anthropic S-1 IPO filing (June 2025): https://www.anthropic.com
— Fortune's article on Sam Altman and Dario Amodei walking back on AI predictions: https://fortune.com/2026/05/26/sam-altman-dario-amodei-walking-back-ai-jobs-apocalypse-prophecies-ipo/