AcceleratingBiz®

Growth CEO’s AI Briefing · Edition #20

Settling Patterns

Monday, May 4, 2026 · 3 Developments · 2 Perspectives · 2 Case Studies · 1 Resource

Designed by AcceleratingBiz®. Curated, synthesized, and produced by Claude agents. AI-assisted production means occasional errors may appear.

After several weeks of breathless announcements, the picture this weekend is quieter and more useful: the AI economy is starting to settle into operational patterns — on procurement, pricing, and how incumbents actually grow. Three of these items resolve open questions from prior editions; the rest reframe how to think about the next 12 months.

Two of these — the Pentagon resolution and Satya's per-seat comment — are worth a leadership team conversation this week.

Developments

Legal & RegulatoryDeveloping

May 1: Pentagon signs classified-network AI deals with eight vendors, formally excludes Anthropic

Building on the White House–Anthropic standoff covered in Friday's What to Watch, the DoD on May 1 announced agreements with SpaceX, OpenAI, Google, Microsoft, AWS, Nvidia, Oracle, and Reflection AI to deploy their models inside IL6 and IL7 networks (secret and top-secret). Anthropic is not on the list. The Pentagon's stated rationale is to avoid "vendor lock," widely read as a reference to Claude's prior dominance via Palantir's Maven platform. Onboarding time for new vendors is reportedly down from 18 months to under three.

So what

For portfolio companies serving federal customers, the procurement reset is real and now operational, not theoretical. The strategic message for any growth company building on Anthropic infrastructure: federal classified work is closed for now, but the injunction remains active and Trump's "shaping up" remark suggests the door is not bolted. Update federal-account roadmaps to reflect a multi-vendor stack as the realistic baseline; treat single-vendor strategies for any classified-adjacent buyer as a procurement risk worth flagging in board materials.

Confidence: Established fact — announcement is on the record; vendor list and exclusion confirmed by Reuters, CNN, WSJ, and the DoD release.

Go deeper CNN: Pentagon strikes deals with 8 Big Tech companies after shunning Anthropic · The Rundown AI

TechnologyDeveloping

May 1: OpenAI completes Codex repositioning — from coding tool to general-purpose work agent

Building on the Codex platform expansion held from Friday's edition, OpenAI on May 1 made the repositioning explicit through coordinated messaging from Altman, Brockman, and the Codex engineering lead: the product now connects to Slack, Google Drive, Gmail, and calendar, runs on Mac with native UI affordances, and is being marketed to non-coders. The Deep View characterizes this as the staged rollout of OpenAI's superapp strategy — Codex, ChatGPT, and Atlas browser converging into one surface. The catch-up move is also obvious: Anthropic shipped Claude Cowork in January with a similar feature set.

So what

The agent-surface war is now clearly a two-horse race for knowledge-work distribution: Cowork on the Anthropic side, Codex on the OpenAI side, both trying to own the "daily AI assistant" layer between user and apps. For SaaS leadership teams, the question is not which model to standardize on but how the product surfaces in both. Get specific about what your product looks like as a tool inside a Codex or Cowork workflow — not as a destination app the user remembers to open. The connectors and MCP servers you ship in the next two quarters are the equivalent of mobile-app submissions in 2009.

Hype check

OpenAI's framing of this as a strategic pivot obscures that the substantive feature set is a deliberate copy of Claude Cowork shipped four months earlier; the news is OpenAI's marketing posture, not a product-capability leap.

Confidence: Established fact — corroborated across three independent sources and OpenAI's own posts.

Go deeper The Deep View: OpenAI turns Codex into a daily work agent · Superhuman AI

Technology

May 4: UK's NCSC issues unusual "patch wave" warning as AI-discovered Linux flaw goes public

The UK's National Cyber Security Centre published guidance on May 4 warning that AI-driven vulnerability discovery is now operating at a scale and speed the industry's patching infrastructure was not designed for; the agency advises organizations to assume incoming updates will be critical severity and prioritize internet-facing systems first. Context: Anthropic's Mythos found 271 Firefox zero-days and over 2,000 unknown flaws in seven weeks of testing; over 99% remain unpatched. Last week, Theori's AI scan of Linux's cryptographic code surfaced "Copy Fail," a one-hour-to-find flaw granting root access on every major Linux distribution shipped since 2017.

So what

The window between "flaw discovered" and "flaw exploited" is collapsing from weeks to hours, and most growth companies' patch cadence assumes the older timeline. Ask your CISO three questions this week: how many of our internet-facing systems can be patched within 24 hours of a critical CVE, what legacy systems can no longer be patched at all (and what is the migration plan), and how would we know within hours rather than days if one of our vendors had been breached. This is also a customer-conversation issue: enterprise buyers will start asking suppliers for evidence of accelerated patch SLAs.

Confidence: Established fact for the NCSC guidance and the Copy Fail vulnerability; informed perspective on the broader timing implication.

Go deeper NCSC: Prepare for a vulnerability patch wave · The Neuron

Perspectives

Business

May 1: Satya Nadella signals the death of per-seat SaaS pricing on Microsoft's Q3 call

In a quote Jamin Ball flagged in his May 1 Clouded Judgement post, Satya Nadella told earnings analysts: "The basic transformation of any per-user business of ours — whether it is productivity, coding, or security — will become a per-user and usage business. That is the best way to think about it." Ball's read: Microsoft, the company with the most installed-base risk from messing with the seat model, has just signaled it is moving to a hybrid where the seat becomes a packaging mechanism for prepaid consumption, with usage above the cap priced per token, per agent action, or per outcome. The shift is structural, not promotional.

So what

If Microsoft is doing this, every SaaS CFO and CRO needs a defensible answer to two questions over the next two quarters: how do we model revenue when AI usage on our platform is non-linear with seat count, and how do we structure renewals so that customer success in adopting agents does not destroy our growth rate. The companies that move first to a credit/consumption hybrid will set the pricing reference points enterprise buyers carry into every other procurement conversation. The companies that wait will find themselves explaining why their pricing model penalizes the customer for getting more value — a conversation that does not end well.

Confidence: Established fact for the Nadella quote (on Microsoft's earnings transcript); informed perspective on the broader pricing-model implication.

Go deeper Clouded Judgement: The Death of Per-Seat Pricing? · TLDR Founders

Socioeconomic

May 3: Notion's Max Schoening — agency, not skill, is what separates people who thrive in the AI era

In Lenny Rachitsky's May 3 podcast, Notion head of product Max Schoening makes the case that the differentiating attribute among PMs and designers in the AI era is not technical skill (the model can supply that) but agency — the willingness to ship code, prototype in the terminal, throw work away, and start again. He also pushes back on the SaaSpocalypse thesis: software volume has exploded but software quality has not, and the gap is the opportunity. His "tiny core" theory — iPhone multitouch, GitHub pull request, Notion blocks, Dropbox menu-bar icon — argues that great products are still built around one small primitive that is hard to imitate.

So what

For talent reviews this cycle, the right diagnostic question is not "does this person know how to use Cursor or Claude Code" but "does this person actually open the terminal, ship something, and learn from breaking it?" Agency is observable in artifacts, not interviews. Pair this with the Schoening framing in your next product offsite: what is your tiny core, and is the team building outward from it or layering features on top? The SaaSpocalypse pushback is also useful as a counterweight in board conversations — the question is which incumbents have a tiny core worth defending and which have only feature breadth.

Confidence: Informed perspective — Schoening's positioning is directly tied to Notion's commercial interest, but the framing is consistent with patterns visible across multiple AI-forward product orgs.

Go deeper Lenny's Podcast: Max Schoening on agency in the AI era

Case Studies

Product & Technology

April 30: Atlassian and Twilio post breakout Q1 quarters, both AI-attributed

Both companies reported on April 30. Atlassian: $1.79B revenue, up 32% YoY, with Cloud accelerating to 29%; Rovo customers grew ARR at roughly 2x the rate of non-Rovo customers, AI credit usage growing 20%+ month-over-month, and Teamwork Collection customers using ~2x more AI credits. Twilio: $1.41B revenue, up 20% reported and 16% organic — the fastest organic growth since 2022. Voice was up 20% YoY, the highest in 19 quarters, with Conversational Intelligence and Branded Calling each up over 100%. Both management teams attributed the acceleration explicitly to AI agent and voice workloads, not to favorable comps.

So what

This is a useful counterweight to the "AI-native startups will replace legacy SaaS" narrative that has dominated recent Substack discourse. Two incumbents pivoted product, repackaged toward usage, and accelerated — not despite AI, but because of it. The right question for portfolio companies is which of those two motions are actually present: have you shipped a product surface where AI usage drives net-new revenue (Twilio's Conversational Intelligence, Atlassian's Rovo credits), and is your sales motion priced to capture that usage rather than dilute it? If neither, the SaaSpocalypse thesis applies to you specifically, regardless of what happens at the category level.

Hype check

Both companies have commercial reasons to attribute strength to AI on a public earnings call — treat the strategic thesis (AI is driving expansion) as well-supported by the disclosed metrics, but treat the precise causal share as management framing rather than independently verified.

Confidence: Established fact for the disclosed financials and management commentary; informed perspective on the broader incumbent-vs-startup implication.

Go deeper SaaStr: Atlassian and Twilio crush the quarter · Superhuman AI

Product & Technology

May 4: Stripe's Protodash — what an internal AI prototyping tool looks like done well

On the May 4 How I AI episode, Stripe design manager Owen Williams describes Protodash, an internal prototyping platform he built that lets designers and PMs turn Stripe's design system into clickable, production-quality prototypes. The architecture: Stripe's Sail design system bundled with React components, an MCP server connecting Cursor to the design library, and a stack of Cursor rules teaching the AI when to query Sail rather than imagine the components. Generic AI prototyping tools (v0, Tailwind-based outputs) produce what Williams calls "blurple slop" — Stripe-shaped but visibly off. Equal numbers of PMs and designers now use the tool; conversations have shifted from "should we staff a designer" to "here's the working prototype, let's discuss."

So what

The transferable lesson is not "build your own Protodash" — most growth companies should not. It is the architectural pattern: take a proprietary asset (design system, brand guidelines, customer-data schema, internal documentation) and wire it to coding agents through an MCP server, then write Cursor or Claude rules that force the agent to query rather than improvise. This pattern is replicable for sales decks, customer-success playbooks, contract templates, support macros, and any other domain where your house style is what makes the output good. Ask one product or design lead to ship a similar internal tool this quarter and treat it as a forcing function for documenting tribal knowledge.

Confidence: Established fact for the case study details as described by Williams; informed perspective on the transferable pattern.

Go deeper How I AI: Demos not memos — Stripe's internal AI prototyping tool

Resources

Reference

Mar Hershenson Miura-Ko's L0–L5 AI adoption levels framework

Floodgate's Miura-Ko adapts the autonomous-vehicle six-level autonomy framework to AI adoption inside companies. L1 is personal-productivity use of AI by individuals; L2 is functional silos (a marketing team has its own AI workflow); L3 is agents acting across CRM, code, and ticketing systems through MCP and shared skills authored by non-engineers; L4 is a compounding operating system where finance staff and AEs ship production internal tools and policy-driven agents handle issues with human review only at merge; L5 (still hypothetical) has the system noticing problems, acting within delegated authority, and updating shared memory without a human starting the loop. Her assessment: most companies that describe themselves as "AI-forward" are still at L1 or L2.

So what

This is a usable diagnostic to take into a leadership team or board conversation this month: where are we actually, by function, on the L0–L5 scale? The framework's value is forcing honest answers. A CEO who claims to be "AI-first" but whose finance team is at L1 (people pasting into ChatGPT) and whose product team is at L3 (agents in the workflow) is not at one level — they have a portfolio of levels, and the gap between them is the actual adoption work. Use the framework to map by function, identify the lowest-rated functions that matter most, and assign one accountable owner per function for the next two-quarter level shift.

Confidence: Informed perspective — the framework is one investor's lens, but it is internally coherent and useful as a diagnostic regardless of whether the level definitions are universally adopted.

Go deeper TLDR Founders: Everyone wants to be AI-pilled. Most companies are still Level 1.

Sources in this edition: The Rundown AI, The Neuron, The Deep View, Superhuman AI, Lenny's Newsletter, TLDR Founders, Clouded Judgement (Substack), SaaStr

← Latest Editions and Search