The Deployment Era Has Started - PM Status Report, 18th May 2026
- The Enterprise AI Arms Race Is Now a Services Business
- Claude for Small Business - Pre-Built Workflows, Not Just a Connection
- Codex on Mobile - and What Async Really Means
- Google Reframes Android as an Intelligence System
- What This Means for Project Environments
No new flagship models this week. What shipped instead: a $4 billion company built specifically to embed AI engineers inside enterprises, a consulting firm rolling out Claude to 364,000 people, and a coding agent that you can supervise from your phone. The interesting question isn’t what AI can do. It’s how organisations are actually deploying it - and what that means for the people running the work.
The Enterprise AI Arms Race Is Now a Services Business
OpenAI launched its Deployment Company on 11 May - a majority-owned subsidiary with over $4 billion in initial capital, 19 partners including Bain, McKinsey, Capgemini, and TPG, and roughly 150 engineers acquired from day one through the Tomoro acquisition. The message was explicit: software subscriptions aren’t enough to capture enterprise value. You need people inside the organisation.
Anthropic’s answer came three days later. Its expanded PwC alliance will train and certify 30,000 U.S. professionals on Claude, with a rollout toward PwC’s full 364,000-person global workforce. The numbers cited aren’t benchmarks - they’re real deployments: insurance underwriting compressed from ten weeks to ten days, a COBOL modernisation programme running on time and under budget at four times the original scope, security incident response cut from hours to minutes. Anthropic’s enterprise market share now edges OpenAI’s - 34.4% versus 32.3% - driven largely by Claude Code adoption.
The Indian IT services sector noticed. The Nifty IT index fell 3.7% on OpenAI’s announcement. Tata Consultancy Services and Infosys both dropped between 2.5% and 4.8%. Those firms have built large businesses around the work of connecting software to enterprise systems. That work is now being offered by the same company that builds the software.
Claude for Small Business - Pre-Built Workflows, Not Just a Connection
Anthropic’s Claude for Small Business launch on 13 May connects Claude directly to QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365 - not through a generic API, but with pre-built workflows for payroll, invoicing, monthly close, and campaigns. The launch customer, a drinkware manufacturer in Oklahoma, summed it up simply: “Hours of looking at stuff that doesn’t matter are gone.”
The design principle is worth noting. Rather than asking users to figure out how AI connects to their tools, Anthropic pre-built the integration and defined the workflow. The AI arrives knowing what it’s there to do.
Codex on Mobile - and What Async Really Means
OpenAI shipped Codex into the ChatGPT iOS and Android apps on 15 May. It’s not about coding on your phone. It’s about running a long agent task on a remote machine and managing it from wherever you happen to be - reviewing progress, approving decisions, redirecting effort mid-commute.
Developers check in on a diff before lunch, approve a shell command on the train, review test results at the end of the day. Four million weekly active users. HIPAA compliance for enterprise workspaces.
Google Reframes Android as an Intelligence System
Google’s Android Show on 12 May introduced Gemini Intelligence - not as an app, but as the operating layer of the device. The framing from Android’s VP was direct: “We’re transitioning from an operating system to an intelligence system.”
In practice that means multi-step automation triggered by context, not commands. A user hovers over a grocery list in their notes app and Gemini parses it, opens a delivery app, and populates a cart. A photo of a travel brochure triggers a search for matching tours. Create My Widget lets users describe a custom home-screen tool in plain language and have it built on the spot.
Rolling out first on Samsung Galaxy S26 and Google Pixel 10 this summer, with watches, cars, and Googlebook laptops to follow.
Gemini Intelligence builds on Google’s flagship model capabilities that have been active in the API since earlier this year. Gemini 3.1 Pro supports a 1-million-token context window and a thinking_level parameter - letting developers tune the balance between speed, cost, and reasoning depth per request. The Android Intelligence layer makes this accessible at the device level. For project applications handling large document sets, full programme archives, or multi-stream stakeholder communications, that combination of context depth and adjustable reasoning remains one of the more practically useful things Google has shipped this year.
What This Means for Project Environments
Three things to carry into your next planning conversation.
AI deployment is now a delivery workstream. The enterprise-scale rollouts this week - PwC, Netflix, Simple Modern - share one thing: someone designed the workflow before the AI touched it. The delivery improvements came from clear task definition, structured outputs, and connected data. That’s project work. If your organisation is planning an AI implementation, treat it like any other systems integration: scope it, resource it, define what “done” means.
Async supervision is a new PM skill. Codex on mobile models a pattern that will become standard across more AI-enabled tasks. The PM’s role isn’t to watch the agent work. It’s to define the checkpoints, review the decision points, and interpret the outputs. Building that discipline now - on lower-stakes tasks - positions you well before it becomes an expectation on a governance-critical workstream.
Context window size changes what you can analyse in one pass. Gemini 3.1 Pro at one million tokens means your entire programme archive - all your risk logs, stakeholder updates, change requests, and post-mortems from the past 12 months - can sit in context simultaneously. That’s not a future capability. It’s available in the API today. A programme retrospective that used to require a workshop can now be structured as a single, deeply-informed analysis. The question is who designs the prompt and owns the output.
All three of those points share an underlying logic: the value comes from design, not from access. Having a one-million-token model, a connected app, or a mobile-supervised agent doesn’t produce results on its own. What produces results is knowing what task to run, what output to expect, and which decisions to keep. That’s always been the PM’s job. It’s just got more surface area.
Frequently Asked Questions
What is OpenAI’s Deployment Company and why does it matter for enterprise projects? OpenAI’s Deployment Company is a $4B subsidiary that embeds specialist engineers directly inside partner organisations to build and integrate AI into existing systems. It matters for project teams because it signals AI is shifting from a software subscription to a professional-services model - similar to how cloud migration required dedicated implementation teams in its early years.
How is Anthropic’s PwC rollout relevant to project management? PwC is rolling out Claude across 364,000 professionals, with real results already cited: insurance underwriting compressed from ten weeks to ten days, security incident response cut from hours to minutes. It’s the most concrete published benchmark of what enterprise-scale AI deployment looks like in professional services right now.
What does Codex on mobile mean for project managers specifically? It demonstrates async AI supervision - running a long-running agent task remotely and checking in to approve or redirect at key decision points. The PM skill it points to is checkpoint design: knowing which decisions in an AI-assisted workflow require human review and which can proceed without it.
Is Gemini 3.1 Pro’s 1-million-token context window useful for project work?
Yes - particularly for retrospectives, programme-level analysis, or synthesising large document sets. One million tokens can hold roughly 750,000 words of context, which covers most programme archives. The thinking_level parameter lets you control cost versus reasoning depth, making it practical for exploratory analysis, not just quick lookups.
If your organisation launched an AI deployment programme tomorrow, what would the project charter look like - and who would own it?
Yes, AI helped me to write this :)