The platform shipped the badge.
Four platform vendors stopped pitching the agent this fortnight and started shipping the badge it has to wear.
Every Sunday I look at the press release a vendor would have written about its own agent a year ago. This fortnight four of them wrote about the agent’s badge instead.
For a year the platform vendors competed on whose agent was smartest. Apple’s was Apple Intelligence. Snowflake’s was Cortex Code. Workday’s was Illuminate. Databricks’s was Mosaic. The pitch was the agent. This fortnight the pitch quietly pivoted. The agent stopped being the product. The contract the agent has to honor — its identity, its scope, its observability surface — became the product.
The CIO noticed first. The buyer who refused to approve four agents from four vendors will approve one badge service from a vendor she already trusts. The platform that ships the badge gets every agent the customer ever runs.
What’s actually shipping this week
1. Apple shipped a Foundation Models protocol that calls Claude or Gemini through one Swift API. WWDC 2026, announced June 9. The LanguageModel protocol is a public Swift interface any cloud provider can implement to expose the same inference surface. Foundation Models added image input, server-side third-party model calls — Claude, Gemini, and others through the same Swift API — and a Dynamic Profiles system for multi-agent workflows. Free to any developer with fewer than two million first-time App Store downloads. The framework goes open source later this summer. Apple ceded the model. Apple captured the API contract every iOS agent will route through.
2. Snowflake made AI Agent Identity GA at the Summit on June 3. Snowflake Summit 2026, June 1-4 at Moscone, 20,000 attendees, 26 new product capabilities across six domains. Every agent that touches Snowflake now needs cryptographic identity, per-agent RBAC, and a complete audit trail at the tool-call level. Cortex Code — rebranded CoCo at the Summit — grew from a February launch to 7,100 accounts and became the fastest-growing product in company history. Snowflake Intelligence was rebranded CoWork on the same stage. The substrate vendor stopped selling the agent. The substrate vendor started selling the badge office.
3. Workday made the Agent System of Record GA in the Q1 FY27 print on May 21. Total revenue $2.542B, up 13% year over year. Subscription revenue $2.35B, up 14.3%. Operating-margin guide raised to 30.5% from 30%. The customer count on Workday’s agent products more than doubled quarter over quarter — 4,000+ customers running at least one Illuminate agent, ACV from agentic AI products up more than 200% year over year. The Agent System of Record went GA: one directory for every agent the enterprise runs, including agents Workday did not build. Workday became the HR system for non-humans.
4. Databricks opened the Data + AI Summit with Microsoft and OpenAI on the same stage. June 15-18 at Moscone, more than 30,000 attendees. Lakebase — the operational Postgres surface the agents write back to — went GA. Satya Nadella and Greg Brockman both keynoted the same morning. The substrate vendor that competes with Azure and partners with OpenAI hosted the CEO of Azure and the president of OpenAI on a single stage. The competitive model providers showed up because that is where the badging is going to be issued.
What I’d ship in your app this week
The badge moved this fortnight. Two small tools your team can ship in two weeks so the agent that calls your product can be recognized, scoped, and audited from the first request.
Feature one: the agent identity column, on the existing users table. Add a principal_type field — human, agent, or service — to the row you already create for every API key. Backfill with human; require the next sign-up flow to set the value explicitly. Every authorization check now knows whether the caller is a person or an agent acting for one. The customer admin sees how many agents have started routing through your product before the CISO asks for the badge.
- Shape. One column added to the users table. One enum check at the auth middleware. No new service.
- Data shape. Per (user_id, principal_type, owner_user_id, scope_list, created_at). The owner is the human responsible for the agent's actions.
- Latency budget. Sub-millisecond at the middleware. No change to the request budget.
- Cost ceiling. Zero compute spend. The infrastructure is one schema change and a backfill.
- Eval. 50 paired requests reviewed monthly by the security PM — a human request and an agent-typed request, identical payload — scored on "would the audit log let the CISO tell them apart."
- Instrumentation. Log the principal_type on every API call. Alert when an agent's call rate crosses five times the median for its scope group.
- Two weeks in. The admin surface on at least three tenants shows an agent count greater than zero. If it doesn't, the customer hasn't shipped any agents that touch you yet — keep the column anyway; you'll need it the day they do.
Feature two: the agent intent log, on the same API call. When an agent calls your endpoint, capture the upstream user prompt alongside the API parameters. One optional header. The customer’s CISO wants to know not just what the agent did, but what the human asked it to do. The intent log is the artifact that answers that question in the post-incident review.
- Shape. One optional header —
X-Agent-Intent— carrying the upstream prompt. Logged with the call. No LLM call to render or validate. - Data shape. Per (call_id, principal_type, intent_text, retrieval_context, output_summary). Stored alongside the existing audit row, queryable by tenant admin.
- Latency budget. None on the call path. Async write to the log.
- Cost ceiling. Storage only — under $30 per month for a 5,000-seat tenant. Existing call-log infrastructure.
- Eval. 40 paired incidents per quarter — agent action plus intent — reviewed by a senior support engineer, scored on "could the customer have prevented this if they'd seen the intent in real time."
- Instrumentation. Surface the intent in the audit-log UI. Make it copyable. Make it searchable.
- Two weeks in. Three customer admins have used the intent search in a real incident review. If they haven't, the intent field is empty — work backward through the integration to find which agent in the customer's stack is dropping it.
Both ship in two weeks with the team you already have. Both put a structured object between the agent in the customer’s stack and the row in your database. The platform vendors moved the badge office onto the keynote stage this fortnight. The vendor that issues the badge gets every agent the customer ever runs. The vendor that doesn’t will be the agent’s third-party API.
Sources
- Apple Outlines Major AI and Developer Tool Updates at 2026 Platforms State of the Union — MacRumors, 2026-06-09
- WWDC 2026 Developer Tools: Foundation Models Now Swaps AI Providers Without Code Changes — TechTimes, 2026-06-09
- WWDC 2026: Everything announced on Siri AI, iOS 27, Apple Intelligence, and more — TechCrunch, 2026-06-09
- Snowflake Summit 2026: All the Announcements and What They Mean — Atlan, 2026-06
- Snowflake Summit 2026 recap: rebrands, acquisition and 25+ announcements — Flexera, 2026-06
- Snowflake Cortex Code is now called CoCo and performs autonomous development tasks — Techzine Global, 2026-06
- Snowflake Summit 2026: Four Infrastructure Bets That Determine Whether the Agentic Enterprise Delivers — Futurum Group, 2026-06
- Workday — Q1 FY27 results, 8-K (May 21, 2026) — Workday / SEC, 2026-05-21
- Workday Illuminate Expands with New AI Agents for HR, Finance, and Industry — Workday Newsroom
- Workday Q1 FY27 earnings: Revenue, margins, and the Flex Credits inflection to watch — TIKR, 2026-05-21
- Databricks Announces 2026 Data + AI Summit Keynote Lineup and Programming for the World’s Largest Data and AI Conference — Databricks newsroom, 2026-06-02
- Databricks sets keynote lineup for Data + AI Summit featuring Microsoft CEO Nadella, OpenAI co-founder Brockman — Sahm Capital, 2026-06-03
- Databricks Data + AI Summit 2026: What Builders Need to Know Before June 15 — ChatForest, 2026-06
- Towards a Science of AI Agent Reliability — Rabanser et al., Princeton (v3) — arXiv 2602.16666v3, 2026-06-02
Send me an email and we will talk. If something here landed close to what you're working on, the door is open. No calendar funnel, no pitch deck — I read every note that comes in.
Doing the work rather than deciding what to build? Crafting is the column for that chair.
← All Signal issues · Drafted with Claude · Edited by Paul Brown