Day 59 · The Agentic Enterprise Buyer's

1. The five stacks in one sentence each

Microsoft Agent 365 (GA May 1, $15/user standalone or $99/user inside M365 E7) is the enterprise control plane -- centralised agent registry inside Microsoft Admin Center, Entra workload identity for agents, real-time anomaly scoring, one-click kill switch and automatic EU AI Act Annex III audit trail generation. If you already run M365 + Entra + Defender, Agent 365 is the lowest-friction governance answer on the

Claude Cowork + Claude Code + Claude Managed Agents (Anthropic, $44B ARR, 4× enterprise share growth vs May 2025) is the developer + skills layer + regulated vertical play. 10 prebuilt finance agents (May 5), 15 ready-to-run Claude for Small Business workflows (May 13), Microsoft 365 add-ins for Excel/PowerPoint/Word/Outlook so context carries between apps, 75+ official Claude connectors, Agent Skills as an open standard (adopted by Microsoft, OpenAI, Atlassian, Figma, Canva, Stripe, Notion, Zapier). Where Agent 365 controls the agent, Cowork ships the agent. OpenAI Codex + Dell AI Factory (May 19 partnership, 4M+ weekly Codex users, 5,000+ Dell AI Factory customers) is the regulated industry counter-move. Codex now plugs into the Dell AI Data Platform on-premises -- directly aimed at UK and EU financial services, healthcare and government buyers who cannot send data to public cloud. Codex extends beyond code into context-gathering, reporting, lead

Google Workspace + Gemini Spark + Antigravity 2.0 + Gemini Enterprise Agent Platform is the consumer + Workspace surface play. Spark runs always-on in the Gemini app for Google AI Ultra subscribers, taps Gmail/Calendar/Drive/Docs/Maps, with a separate Skills layer and Gemini 3.5 Flash (default model, ~1/3 cost of rival frontier models). Antigravity 2.0 is Google's Claude-Code-equivalent for developers -- together, Google plays both consumer OS surface and developer + skills layers. Salesforce Agentforce 3.0 (18,500+ customers, $500M ARR, Flex Credits $500/100K, 20 credits per action) is the CRM-native vertical play. Self-Healing Workflows (cosine<0.7 triggers context reset, Day 24), A2A + MCP native, T1-T4 kill switch, Microsoft 365 add-ins for finance -- if your team lives in Salesforce, Agentforce wins on data gravity. Outside Salesforce, the per-action economics get expensive for

2. Buyer's matrix -- five evaluation lenses

Control plane. Microsoft Agent 365 leads (centralised registry + Entra identity + Defender hooks + Admin Center kill switch). Anthropic Cowork is the new contender (Agent View + audit trail + scoped MCP). Salesforce Agentforce is best inside Salesforce. Google has Gemini Enterprise Agent Platform but it's less mature for non-Workspace agents. OpenAI + Dell is best if your control plane is Dell AI Factory itself. Skills ecosystem. Anthropic leads decisively -- Agent Skills as an open standard already adopted by Microsoft, OpenAI, Atlassian, Figma, Canva, Stripe, Notion and Zapier; 75+ official connectors; Skills Marketplace inside Cowork; Plugins for Code. Google's Spark Skills layer is announced but not open yet. Microsoft Copilot Studio + APM (apm.yml) is the runner-up. OpenAI GPT Store + MCP Hubs are useful but fragmented. Salesforce keeps the Agentforce skill catalogue inside its own walled garden. Governance evidence. Microsoft Agent 365 (Annex III audit trail by construction) and Anthropic Cowork (NIST CAISI WORM + SPIFFE-style identity + kill switch T1-T4) tie for first. Salesforce Agentforce 3.0 ships Self-Healing + T1-T4. OpenAI + Dell inherits Dell AI Factory governance (strong, but the Codex layer doesn't publish its own EU AI Act mapping yet). Google Antigravity 2.0 currently has no product-specific compliance documentation -- GCP certifications don't automatically extend. For Annex III shops, this is the single biggest differentiator on May 21. Distribution and surface. Google wins on raw reach (Spark inside Gemini app for 3B+ Android users, Workspace inside Gmail/Calendar/Drive). Meta's Muse Spark reaches the same 3B+ on WhatsApp/Instagram/Facebook/Messenger. Microsoft Agent 365 reaches every M365 E3+ seat (~400M+ paid). Anthropic Cowork reaches the high-value 'developer + knowledge worker' tier. OpenAI + Dell reaches the regulated enterprise. Salesforce Agentforce reaches every Sales Cloud + Service Cloud seat

Pricing per outcome. Microsoft Agent 365 is the new horizontal ceiling at $99/user/month all-in (M365 E7). Salesforce Flex Credits are $500 per 100K, 20 credits per Agentforce action, 30 per Voice action -- predictable for low-volume, expensive at high-volume (200K credits free with Salesforce Foundations). Anthropic Cowork prices on plan + usage credits ($20-$200/mo paid plans for SMB, enterprise negotiated). OpenAI Codex is the cheapest per-developer at consumer pricing but on-prem with Dell is enterprise-quoted. Google Antigravity is $100/mo Ultra (no team tier yet). Outcome: the per-task pricing wave (predicted Day 58) is already showing in Salesforce Flex Credits and AWS AgentCore Payments.

3. The reference enterprise blueprint

The 2026 enterprise agentic stack is no longer a single vendor -- it is a tiered combination. Most Fortune 500 firms running pilots converge on the same shape: a horizontal control plane (Microsoft Agent 365 OR Anthropic Cowork as the registry + governance layer), plus one or two vertical agents tied to systems of record (Salesforce Agentforce for CRM, ServiceNow for IT/HR, Workday for finance, JP Morgan-style custom finance agents), plus a regulated-industry tier (OpenAI Codex + Dell AI Factory) for anything that

Skills are written once, distributed everywhere. The cheapest deliverable in 2026 is a Skill folder (SKILL.md + reference scripts + tests + apm.yml) -- versionable in git, auditable for Annex III, free to publish. Mirror the same Skill as an MCP server, register it with Google A2A, and apply for inclusion in Spark's Skills marketplace when it opens. One artefact, four distribution channels. Identity is per-agent, not per-user. Every counter-move ships some form of SPIFFE/SPIRE-style workload identity -- Microsoft Entra workload identity, AWS Bedrock AgentCore + SVID, Anthropic Cowork agent identity, Salesforce Agentforce per-agent OAuth scopes. KYA (Know Your Agent) is now the standard. Day 54's identity ladder is no longer aspirational -- it's the buyer's checklist.

4. Decision shortcuts -- pick by your starting point

If you are an M365 + Entra shop with 5,000+ knowledge workers and you need to ship something governable in the next 90 days before Aug 2 enforcement: Microsoft Agent 365 (M365 E7 bundle) plus Claude Cowork or Anthropic finance agents as the value layer. Agent 365 handles the audit trail by construction; Cowork brings the workflows. If you are a Salesforce-centric revenue org and your number-one priority is CRM-resident automation: Agentforce 3.0 with Flex Credits, Self-Healing Workflows, A2A + MCP. Add Claude Cowork later for cross-domain knowledge work that escapes the CRM boundary. If you sell into regulated industries (financial services, healthcare, federal/state government, defence) and you have a Dell footprint: OpenAI Codex + Dell AI Factory on-premises (May 19 announcement is the first explicit hybrid play). Layer Claude Cowork on top for the workflows that don't touch the regulated data. If you are a Google Workspace shop and your users live in Gmail/Calendar/Drive/Docs: Gemini Spark (consumer surface) + Gemini Enterprise Agent Platform (control plane) + Antigravity 2.0 (developer tier). Plan Annex III audit packs separately -- Google Antigravity has no product-specific compliance docs yet. If you are an indie developer or a B2B SaaS founder, you don't pick -- you ship into all of them. Build the Skill once, mirror as MCP, register with Google A2A, sell it through Anthropic's Skills Marketplace + Microsoft Copilot Studio + ChatGPT GPT Store + Salesforce AppExchange. Distribution is now

5. Five things breaking this week

Nvidia Q1 FY27 earnings (May 20): data center revenue $75.2B = 92% of total revenue, doubled YoY. Jensen Huang: 'Agentic AI has arrived'. New Vera CPU opens a $200B TAM with every major hyperscaler partnering. Stock slid post-earnings on China export uncertainty, but the agentic compute thesis is now

Google Antigravity 2.0 (May 2026): Google detaches the agent from the IDE completely -- desktop + CLI + SDK + Managed Agents API + Enterprise Agent Platform. Dynamic subagents = Three-Agent Harness as a first-class product feature. Direct Claude Code rival; community benchmarks suggest 12× speed on some workflows. Caveat: no team pricing, no compliance docs. EU AI Act T-73 days. 78% enterprises NOT meaningfully on track per Secure Privacy April survey. Over 50% lack a basic AI inventory; 40% have unclear risk classification. Up to €35M / 7% global turnover for prohibited practices, €15M / 3% for high-risk non-compliance. The Digital Omnibus extension (proposed to push Annex III to Dec 2027) remains a proposal -- prudent compliance still treats August 2026 as binding. Hyperscaler 2026 AI capex hits $710B combined (Amazon $200B, Microsoft $190B, Alphabet $185B, Meta $135B). Nvidia data centre revenue absorbing nearly half of it. The financial proof that agentic distribution + the agentic control plane has displaced 'just running models' as the cash centre. OutSystems research (April): 96% of enterprises now using AI agents in some form, 97% exploring system-wide agentic strategies -- but 94% raising serious concern about AI sprawl, complexity, technical debt, and security risk. Agent 365's central registry + Cowork's audit trail are the answer; the buyer's-guide

Two days after I/O 2026 Google quietly shipped Antigravity 2.0 -- and the AI dev community noticed instantly. Where the original Antigravity was an agent-first IDE built on VS Code, the 2.0 release detaches the agent from the editor completely: desktop app, CLI, SDK, Managed Agents in the Gemini API, and enterprise deployment through the Gemini Enterprise Agent Platform. The killer new primitive is dynamic subagents -- the primary agent spawns focused mini-agents for sub-tasks that operate independently in isolated context windows and report back. This is the Three-Agent Harness pattern (Day 23) made a first-class product feature, and a direct rival to Claude Code's Auto Mode + Routines + Agent View stack. Pricing: included in Google AI Pro, $100/mo Ultra (5× Pro), $200/mo Ultra Premium (20× Pro). Why it matters: Antigravity proves Google is willing to compete on the developer + skills layer too -- not just the consumer OS surface. Spark + Antigravity = Google playing offense on both sides of the agentic stack at once. One open question: no team pricing, no product-specific compliance docs, so Annex III evidence still has to be built on top.

MARKET SIGNAL

The 2026 enterprise agentic buyer's market just split cleanly into four lanes: horizontal control plane (Microsoft Agent 365 wins by default for M365 shops), workflow + skills layer (Anthropic Cowork owns the open Skills standard and the regulated-vertical agent templates), regulated-industry on-prem (OpenAI + Dell AI Factory is now the only credible hyperscaler play for data-residency buyers), and CRM-resident vertical (Salesforce Agentforce 3.0 inside the Salesforce gravity well). Google Antigravity 2.0 is the wildcard -- credible on speed and feature breadth, but its lack of product-specific compliance documentation is a real Annex III blocker with 73 days left. The smart buyer's move is no longer 'pick one' -- it is 'pick the control plane, then pick the value layer, then pick the regulated-industry tier'. Per-seat pricing is collapsing into per-task and per-outcome at the same time.