Three things happened this week that, taken separately, look like ordinary product news. Taken together, they represent something more significant: the agentic playbook for enterprise AI has stopped being a vision statement and started being a vendor pitch deck.
IBM Think: “AI operating model” becomes a real phrase
IBM’s annual Think conference delivered the most explicit articulation of an enterprise AI operating model we’ve seen from a major vendor. The announcement wasn’t about a new model or a headline benchmark. It was about architecture — specifically, how enterprises wire agents into their operational layer rather than bolting them on as a pilot.
Watsonx Orchestrate next-gen is the centerpiece: a multi-agent orchestration platform that sits above the existing enterprise stack and coordinates work across CRM, ERP, and ITSM systems without requiring those systems to be replaced. The framing is deliberate. IBM isn’t selling you on ripping out SAP. It’s selling you on putting an intelligent layer on top of it.
IBM Bob — an agentic developer partner with security controls and cost guardrails built in — is the other notable announcement. The name is a choice, but the product logic is sound: if you’re going to have agents writing and reviewing code inside a Fortune 500, you need audit trails, permission models, and rate controls that your existing security team can actually understand. IBM Bob has those. Most AI coding tools don’t.
What it means: The “AI operating model” framing is going to become table stakes in enterprise AI vendor conversations over the next 12 months. If you can’t explain how your product fits into the operating model — not just the tech stack, but the actual org structure — you’re going to lose deals to vendors who can.
Anthropic plants a flag in financial services
The same week as IBM Think, Anthropic ran its own event in New York focused entirely on financial services. The announcements were surgical: pre-built agent workflows for the largest banks, Claude Opus 4.7 optimized for financial analysis, a Microsoft 365 integration that puts Claude inside the tools bank employees actually use, and a Moody’s data partnership that connects the model to real-time financial signals.
The two-track strategy is interesting. Track one: go deep with the top-ten institutions — custom models, dedicated support, bespoke agent workflows. Track two: a PE-backed joint venture pushing Claude into the mid-market wealth management and commercial banking segment that the big banks don’t serve.
The ARR number that circulated this week — Anthropic’s revenue has now eclipsed OpenAI’s enterprise ARR — is the headline most people led with. We’d push back on making too much of that comparison at this stage. What matters more is the pattern: a frontier model lab is going vertical-first, with dedicated workflows, data partnerships, and a two-tier go-to-market. Vertical AI for finance just stopped being a thesis and started being a market.
What it means: The playbook that worked in legal AI (specialist models, domain data, workflow integrations) is being replicated in finance at speed. If you’re building in adjacent verticals — insurance, real estate, private equity back-office — watch what Anthropic just did and assume a competitor will do it to your segment within 18 months.
Forward-deployed engineering becomes the new default
The third move was arguably the most structurally important. ServiceNow and Accenture announced an FDE (forward-deployed engineering) program this week: 300+ pre-built agent skills, plus dedicated engineers embedded with the customer to handle the last-mile implementation work that no software product can fully automate.
Cognizant launched a parallel program on May 7th — “Secure AI Services” — with similar architecture: pre-built agent modules, a dedicated deployment team, and a focus on industries where data security and auditability are non-negotiable.
Palantir built its entire go-to-market around FDE a decade ago and was mocked for how expensive it was to operate. The rest of the enterprise software industry is now copying it, because it turns out the bottleneck to enterprise AI adoption isn’t the model — it’s the integration, change management, and ongoing tuning work that happens after the sale.
What it means: FDE programs are expensive to run at scale, which means smaller vendors can’t afford to do it properly. The differentiation is going to come from companies that can automate the deployment muscle — better tooling for environment setup, faster agent skill transfer, and AI-assisted change management. That’s an underserved market today.
The through-line
What connects these three announcements is a single frame: enterprise AI has moved from “what is this?” to “how do we operate it?” That’s a very different question, and it has very different answers. Operating model, vertical workflow, deployment muscle. That’s the actual product now.
Vendors who are still selling on benchmark performance are about to find out the hard way.