AI Brief — what's new in enterprise AI
A curated digest of what's happening in enterprise AI — filtered for what's actually practical for Scotts, not the hype cycle.
How this is filtered
We pull from a small set of trusted sources (AI Daily Brief, vendor enterprise blogs, select enterprise-AI commentary), look at the last 14 days, and shortlist 5 items per fortnight: 3 Gemini-stack picks + 2 broader enterprise AI picks. The Gemini bias reflects what our associates use company-wide today; the broader picks keep us aware of the landscape (Claude, governance, agentic patterns, EU AI Act timeline). Every item gets a "why this matters for Scotts" line — that's the part that does the real work.
Two fortnights stacked. The current digest sits at the top; the prior fortnight's stays visible below it so the trajectory is legible (e.g., last fortnight's Shared Gems → this fortnight's Shared conversations). Older fortnights roll off the page each cycle.
What gets boosted:
- Gemini stack — Gemini app/Enterprise/Workspace, NotebookLM, Gems, Vertex AI updates that affect our daily tools
- Anthropic / Claude enterprise news (we move to Claude Enterprise in FY27)
- Governance, policy, audit, compliance for enterprise AI
- COE patterns, change management, adoption frameworks
- Agentic deployment patterns and tooling
- Cost / ROI signal (token economics, deployment patterns)
What gets suppressed:
- Raw research papers and model benchmark wars
- Consumer features that don't map to enterprise use
- Hype-cycle takes without a concrete enterprise hook
Cadence: Refreshed every Wednesday (the day before office hours).
Top 5 — week of 2026-06-17
Covers 2026-06-03 → 2026-06-17. Anchored on Google I/O 2026 (May 19–20) drops landing in Gemini Enterprise, and the EU AI Act Code of Practice published June 10. Hand-curated this fortnight; the scheduled filter wires up before the next digest.
NotebookLM sources now stay current automatically — no more manual re-syncing
The news. Google rolled out automatic Drive syncing in NotebookLM starting May 26. Google Docs, Sheets, and Slides files added as notebook sources now update in the background whenever the originals change in Drive — no manual refresh required. Permissions are enforced automatically too: if a user's access to a source file is revoked in Drive, that source becomes unavailable in the notebook immediately.
Why this matters for Scotts. Any NotebookLM notebook built around living documents — seasonal planning decks, SOPs, product specs — was silently going stale the moment the source file changed. That friction was the main barrier called out in the NotebookLM Enterprise intake card. This update removes it: notebooks built for team knowledge management now reflect current content without any upkeep overhead, and the permission-inheritance model means enterprise data governance stays intact.
Gemini 3.5 Flash is now the locked-in default across Gemini Enterprise
The news. As of June 9, Gemini 3.5 Flash became the non-disableable default model in the Gemini Enterprise app for all regions (Global, US, EU). The admin feature-management toggle that previously allowed orgs to opt out was removed on June 16. Flash is also now available in the Gemini Enterprise Agent Designer for building and running agents.
Why this matters for Scotts. Any agent or assistant built in Gemini Enterprise automatically runs on the faster, more capable Flash model — existing workflows get a silent capability upgrade with no rebuild required. Worth a spot check of any agent prompts currently in the Production Planning Tool intake card that were calibrated for an earlier model tier, as output style and reasoning depth may shift noticeably.
Gemini CLI sunsets tomorrow — enterprise licences are safe, individual users must act now
The news. On June 18 the Gemini CLI stops serving requests for all Google AI Pro, Ultra, and free-tier Gemini Code Assist for individuals accounts, with no grace period. The replacement is Antigravity CLI (agy), a Go-based rewrite with multi-agent and background-workflow support. Organisations on Gemini Code Assist Standard or Enterprise licences retain uninterrupted access and are not forced to migrate — but the new CLI is now the canonical developer surface for Antigravity 2.0.
Why this matters for Scotts. Any associate or team using the gemini command in personal scripting, CI pipelines, or IDE extensions under a personal licence will see an outage at midnight. Enterprise licence holders (the likely profile for most Hub builders) are carved out, but any automation referencing gemini should be audited now regardless. This intersects directly with the Agentic Workflow Builder intake card — pipelines designed on Antigravity 2.0 CLI will use agy syntax from day one; the Hub should update any sample scripts accordingly.
Next two: outside the Gemini stack — Claude's enterprise agent controls (directly relevant to FY27 platform transition) and the EU AI Act August 2 hard deadline for agentic governance.
Claude Managed Agents now run in your own infrastructure — with private MCP
The news. Anthropic shipped two enterprise capabilities for Claude Managed Agents: self-hosted sandboxes (public beta) and private MCP server connectivity. Tool execution can now run on Scotts-controlled infrastructure (or managed providers like Cloudflare, Vercel, or Modal) while the agent orchestration loop stays on Anthropic's side. Both the execution environment and the services agents call can now operate inside your enterprise boundary.
Why this matters for Scotts. The FY27 move to Claude Enterprise is explicitly named in our strategy frame. This update resolves the primary enterprise objection to agentic Claude — data leaving the org perimeter — and means the Hub's Blocked Stock Workflow or any SAP-connected agent concept can be architectured with tool calls that never leave Scotts infrastructure. Worth surfacing to the AI Governance Committee before the FY27 platform decision is locked.
EU AI Act high-risk obligations activate in 46 days — agentic systems are in scope
The news. Articles 8–17, 26, 27, and 73 of the EU AI Act — covering risk management systems, human oversight, audit trails, and deployer obligations for high-risk AI — become mandatory on August 2, 2026. Penalties for non-compliance run up to €35 million or 7% of global revenue. Agentic AI systems that automate decisions involving personal data, workforce processes, or supply-chain operations are likely in-scope as high-risk deployers.
Why this matters for Scotts. Any agentic workflow the Hub ships between now and August 2 — whether on Gemini Enterprise, Antigravity 2.0, or the FY27 Claude platform — needs an evidence trail: a control catalog, a compliance matrix, and a risk register per the Act's Article 9. The AI Governance Committee should table a readiness review at the next Office Hours; the 30-60-90 day governance rollout framework cited in current practitioner guidance maps directly to our remaining window.
Top 6 — week of 2026-06-04
Covers 2026-05-19 → 2026-06-04, anchored on Microsoft Build 2026 (June 2 keynote). 4 Gemini-stack picks (the tools everyone uses today) followed by 2 broader picks (Microsoft's full agentic push + the Claude platform piece that touches Bloom). Continuity note: last fortnight's "EU AI Act 10 weeks out" alarm has been overtaken — the May 7 political agreement pushed high-risk AI compliance from 2026-08-02 to 2027-12-02. Hand-curated this fortnight; the scheduled filter wires up before the next digest.
Ask Gemini in Drive can pull from Gmail + Organize My Files goes GA
The news. Two related GA moves push Gemini deeper into Drive. Gmail as a source in Ask Gemini in Drive is now generally available — Workspace users can ground Drive-side Gemini answers in a combined view of email + files + folders. Rollout began 2026-06-03 with up to 15 days for visibility. Separately, Organize My Files in Drive — the Gemini-suggested file-move tool in beta since October 2025 — is also GA, with promotional access to higher per-user limits through 2026-07-15. Both require Workspace smart features enabled.
Why this matters for Scotts. Both items push Drive toward being the grounded surface for everyday associate Gemini use — exactly the direction we need given this week's Trusting AI outputs spotlight. The more an associate can ask Gemini questions anchored to real emails + files instead of free-form chat, the smaller the surface area for fabricated answers. Action: Workspace admin confirms smart features are on; Tier 1 office-hours nudge — "if you want a quick answer about a project, start in Drive — not the Gemini app."
Workspace Updates — Gmail as source in Ask Gemini in Drive → · Workspace Updates — Organize My Files GA →
Gemini conversations + canvases + media are now shareable via Drive — snapshot model
The news. Google extended last fortnight's Shared-Gems story to the conversations themselves. Workspace users on the web can now share chats, Canvases, and generated media from the Gemini app using the standard Drive sharing menu. Shares are snapshots — if a recipient continues the chat, the owner's chat doesn't update. ON by default, governed by existing Drive sharing policies, with a new admin toggle in Admin console → Generative AI → Gemini app.
Why this matters for Scotts. Shared Gems made the template portable. Shared conversations make the work portable. Caitlin's Creative Brief Gem output — the actual brief — can now ride from Marketing to the agency in the same UX as a Google Doc; Tish's Product Knowledge GEMs chat outputs can land in a shared Consumer Services folder. Action: Workspace admin team confirms the toggle is on before 2026-06-03 user rollout; this is also the right week to nudge the Marketing intake authors to start publishing sample Gemini conversations to a shared "AI Hub examples" Drive folder.
Google Workspace Updates — sharing via Drive → · Admin toggle reference →
Ask Gemini in Meet moves to the bottom-left — auto-rollout, no admin action needed
The news. Google moved the Ask Gemini prompt box from a hover-only icon in the top-right to a persistent button in the bottom-left of the Google Meet web interface. Same feature (meeting summaries, action-item extraction, late-joiner catch-up when "Take Notes for Me" is on) — just much more discoverable. Available to Workspace Business and Enterprise Standard/Plus users; if your domain already has Ask Gemini enabled, the new prompt appears automatically.
Why this matters for Scotts. Meet is the universal surface — every associate is in it daily. The single biggest gap with Gemini-in-Meet to date hasn't been capability, it's been that people forget it exists. A persistent bottom-left button changes that: it converts a feature you have to remember into one you see. Expect a real bump in summary + action-item usage across our Tier 1 audience over the next two weeks. Action: Workspace admin team confirms "Take Notes for Me" is on for the licenses that have Ask Gemini; if office-hours wants a 30-second nudge moment, "look at the bottom-left next time you're in Meet" is a natural ask.
Google Workspace Updates — Ask Gemini in Meet more accessible →
Gemini CLI 2026.05 — unified Auto mode, agent registration, stronger sandbox + auth
The news. Google shipped a broad Gemini CLI refresh: Auto mode is now unified across the CLI surface, agent registration and session handling are reworked, and there are improvements to auth, proxy, sandboxing, and model handling — plus new editor and skill integrations. It's the most substantial CLI release since the agent-mode reframe.
Why this matters for Scotts. Bloom is Claude-Agent-SDK-based, so this isn't a direct drop-in — but the patterns matter. Google's CLI now treats agents, sandboxes, and skills as first-class objects; Bloom's skill-library + Cloud Run sandbox design tracks the same direction. For any Scotts engineer prototyping with Gemini Code Assist or building Vertex-side helpers alongside Bloom, this is the version to be on. Action: Bloom team does a 30-min compare against Bloom's skill-library + scratchpad pattern — is there anything in the new CLI's agent-registration story worth borrowing?
Next two: outside the Gemini stack — Microsoft's all-in agentic push + the Claude platform piece that directly touches Bloom.
Microsoft Build 2026 — Agent Mode is the default for Office 365 Copilot, Project Polaris cuts the OpenAI cord
The news. Satya Nadella's framing: "Agents are not just a feature. They are the new operating system for work." Agent Mode is now the default across Word, Excel, PowerPoint, Teams, and Outlook in Office 365 Copilot — rolling out to subscribers in late June 2026 on a consumption-based pricing model (50–75 CCUs per feature request, 2,000 CCUs/team/month free). Microsoft also unveiled Project Polaris, its in-house model replacing GPT-4 Turbo as the default Copilot reasoning engine in August 2026, plus AgentGuard for governance (role-based permissions, DLP, audit), Azure Agent Mesh as a control plane, and Windows Local AI running agents on NPU.
Why this matters for Scotts. We're a Google-stack shop, so the day-one impact is low — but the competitive geometry matters for our strategy frame. Microsoft now has its own answer to Gemini Spark (Agent Mode in productivity apps), its own model independence story (Polaris), and its own governance-as-a-platform pitch (AgentGuard). That changes how vendors will pitch Scotts; expect Microsoft account teams to circle back on Office 365 with a sharper agentic story by Q3. Action: the AI Governance Committee should read the AgentGuard spec — not because we're switching, but because procurement-side agent governance language is converging across vendors and our Bloom governance posture should speak the same dialect.
Windows Forum — Build 2026 recap → · Technobezz — Agent Mode default →
Claude Managed Agents add self-hosted sandboxes + MCP tunnels — agent loop stays Anthropic, tool execution stays ours
The news. Anthropic shipped two features at Code with Claude London that change the deployment model for managed Claude agents. Self-hosted sandboxes (public beta) move tool execution to your own infrastructure — or to a managed provider (Cloudflare, Daytona, Modal, Vercel) — while the agent loop, orchestration, context, and error recovery stay on Anthropic. MCP tunnels (research preview) let agents reach private MCP servers inside your network without inbound firewall rules — a lightweight gateway makes one outbound encrypted connection.
Why this matters for Scotts. This is the most direct hit on the Bloom roadmap question we flagged last fortnight (Antigravity vs. Bloom). Bloom is Claude-Agent-SDK on Cloud Run with a skill library on GCS — Anthropic's new model is "managed loop, your sandbox," which lines up with how Bloom already isolates tool execution per session. MCP tunnels are even more interesting: Scotts has internal data sources (SAP, Snowflake, intranet wikis) that would be candidates for MCP-exposure, and tunnels remove the "publicly exposed endpoint" objection that has slowed those conversations. Reinforces the FY27 Claude Enterprise transition direction. Action: Bloom team reviews the self-hosted-sandbox public beta against our current Cloud Run Job pattern — is there a graceful way to slot in without losing skill-library + scratchpad ergonomics?
Anthropic — Claude Managed Agents updates → · InfoQ — MCP tunnels for private agent access →
Past digests
Older digests are archived chronologically. We don't keep weekly digests around forever — items older than 30 days roll off the front, and only the ones referenced by a Solution Library entry stay linked from the library.