Token Pricing, AI Layoffs, and What Instagram's PM Structure Tells Us

Benedict Evans — Ways to Think About Token Pricing#

Source: https://www.ben-evans.com/benedictevans/2026/7/9/ways-to-think-about-token-pricing Credibility: High (independent analyst, consistently rigorous frameworks, no sponsored content)

What happened: Evans published a fresh framework for thinking about token pricing — how to interpret cost curves, what they signal about competitive dynamics, and how to reason about AI economics as a PM or product strategist. The piece isn't a benchmark comparison or a model ranking. It's a conceptual tool for making sense of the token economy without getting lost in per-million-token math.

Key frameworks:

• Token pricing is not a stable unit of measure — it's a moving target driven by model quality, provider competition, and infrastructure efficiency simultaneously
• The right question isn't "how much does this cost per token?" but "what is the relationship between cost, quality, and the value I'm delivering to users?"
• As prices drop, the relevant constraint shifts from cost to latency, reliability, and differentiation — which changes the build-vs-buy calculus entirely
• Incumbent pricing power is weaker than it looks: substitutability between frontier models is increasing faster than lock-in is being established

Why it matters for PMs: If you're making AI feature investment decisions based on current token costs, you're making a fragile bet. Evans' framework pushes you to ask what happens to your product if token costs fall by 10x in 18 months — does your feature still make sense? Does your pricing model? The "cost is prohibitive" justification for not building something today may evaporate faster than your roadmap assumes. Conversely, if your differentiation is currently built on doing something cheaper than a competitor, that moat is also at risk.

Critical questions:

• If token costs drop dramatically, which of your current AI features become table stakes instead of differentiators?
• How much of your AI product strategy is implicitly betting on current cost levels staying stable?
• Are you tracking latency and reliability curves with the same rigor as cost curves — and should you be?
• What does "quality" mean for your specific use case, and are you measuring it independently of cost?

Action you could take today: Pull up the AI features on your roadmap and mark which ones are gated by cost today. Then ask: if that cost dropped 5x, would you still prioritize them the same way, or would a different feature become the better bet?

Cassie Kozyrkov — What's Really Behind the "AI Layoffs"?#

Source: https://kozyrkov.medium.com/whats-really-behind-the-so-called-ai-layoffs-92f94e126ffd Credibility: High (former Chief Decision Scientist at Google, writes from practitioner experience, published in her own voice)

What happened: Kozyrkov published a piece arguing that what's being called "AI layoffs" is mostly being misdiagnosed. Her take: companies aren't replacing people with AI — they're reorganizing around a productivity step-change, and the workforce effects are more about role redefinition and delayed hiring than mass displacement. The framing matters a lot for how product teams think about their own org design and where to invest in tooling vs. headcount.

Key patterns:

• The "AI layoff" narrative conflates three separate dynamics: efficiency gains reducing marginal headcount needs, companies that over-hired during 2021-2023 now correcting, and genuine automation displacing specific roles
• Productivity unlocked by AI is often absorbed as expanded scope rather than reduced headcount — one person does more, not fewer people do the same
• The more dangerous version isn't replacement but role erosion: tasks get automated, the role shrinks, and the person becomes less valuable without realizing it
• PMs should watch for "productivity theater" — teams that add AI tools but don't restructure work to capture the actual value

Why it matters for PMs: Kozyrkov is pointing at something that directly affects how you plan team capacity and tool investments. If AI tools genuinely expand what a PM or engineer can do, the question isn't "can I reduce headcount?" — it's "am I restructuring the work to capture that expansion?" Teams that add AI tools but keep the same processes, same output expectations, and same review cycles are leaving most of the value on the table. This is also a useful frame for stakeholder conversations about AI ROI: the return often shows up as velocity and scope, not headcount reduction.

Critical questions:

• On your team, when AI tools are adopted, do you actually restructure the work — or just add the tool to the existing process?
• What roles on your team are most at risk of "role erosion" rather than elimination, and are you paying attention to that?
• How do you measure whether AI tooling is actually expanding output vs. just shifting where time gets spent?
• Are you making headcount decisions based on current productivity baselines that will look wrong in 12 months?

Action you could take today: Pick one workflow on your team where AI tools are already being used. Map what the person was doing before vs. what they're doing now. Is the scope of their work actually larger, or did the tool just speed up the same tasks?

Lenny Rachitsky / Adam Mosseri — Product Team Structure in 2026#

Source: https://www.lennysnewsletter.com/p/adam-mosseri-ai-is-a-tailwind-for Credibility: High (Mosseri is the sitting head of Instagram, first-party perspective on how a major consumer product org is structured)

What happened: Adam Mosseri joined Lenny's podcast to talk about how Instagram's product team structure has evolved in 2026. The headline insight: Instagram has formalized a "product staff" role — a senior IC track for product people who stay close to the work without moving into management. Mosseri also talked about how AI has, counterintuitively, been a tailwind for authenticity in product teams rather than a threat to it.

Key patterns:

• The "product staff" role at Instagram fills a gap that many orgs paper over with senior PMs who drift toward management — it's an explicit senior IC track with defined scope and expectations
• Mosseri's take on AI and authenticity: because AI can generate competent generic work, the value of personal judgment, taste, and genuine user empathy goes up, not down
• Hiring criteria have shifted: Instagram now weights "do they actually use our product?" and "do they have a point of view?" more heavily than traditional PM credentials
• Smaller, more senior teams are the pattern — fewer coordinators, more people who can do the actual product thinking

Why it matters for PMs: Two things here worth paying attention to. First, if you're building out a product team and debating whether to promote good individual contributors into management, Mosseri's "product staff" framing is worth stealing. It names something real: some of your best product thinkers will be worse managers, and you need a track for them that isn't "tolerated as a quirky senior PM." Second, the authenticity angle is genuinely useful counter-programming to the anxiety about AI replacing product work. The generic output problem — AI can write a passable PRD — raises the value of specific, defensible, personal product judgment. That's a craft argument for investing in taste, not just process.

Critical questions:

• Does your org have a real senior IC track for PMs, or does everyone eventually get pushed toward management?
• What does "product judgment" or "taste" actually mean on your team, and how do you evaluate it in hiring or promotion?
• If AI handles more of the generative / synthesis work, what does a great PM's day actually look like?
• How does "using the product yourself" hold up as a hiring signal for B2B tools or fintech products where dogfooding is harder?

Action you could take today: If you have a senior PM who's been on the fence about moving into management, think about whether a product staff-style role definition would actually serve them and the team better. What would the job description look like?

Notion — Agents iOS App Launched#

Source: https://www.notion.so/releases/2026-07-08 Credibility: High (first-party changelog)

What happened: Notion shipped the Agents iOS app. This is a dedicated mobile app for Notion's AI agents — separate from the main Notion app — letting users interact with and manage autonomous workflows from their phones.

Why it matters for PMs: Notion is betting that agentic workflows need a distinct mobile surface, not just a feature inside the existing app. That's a product philosophy call worth noting: when your AI features are autonomous enough, they might deserve their own container. The question for any PM building agents is whether the interaction model (check in, review, redirect) is different enough from the core product to warrant separation.

Critical questions: Is Notion's agents-as-separate-app approach a principled UX decision or a workaround for how hard it is to integrate deeply into the existing app?

AWS — Claude Apps Gateway for AWS#

Source: https://aws.amazon.com/blogs/machine-learning/introducing-claude-apps-gateway-for-aws/ Credibility: High (first-party AWS announcement)

What happened: AWS launched the Claude apps gateway — a self-hosted control plane that gives enterprise IT teams a single point of control over access, cost, and policy for Claude Code and Claude Desktop. It runs on top of Amazon Bedrock.

Why it matters for PMs: This is the enterprise AI governance play becoming infrastructure. If you're in a company where Claude Code or Claude Desktop adoption is spreading without IT oversight, this is the product that unblocks formal adoption. The pattern here is important: AI tools spread user-first, then IT has to catch up with controls. AWS just gave IT teams a catch-up mechanism. For PM teams trying to get AI coding tools approved by security, this is a useful reference point for what "enterprise-ready" looks like.