title: AI Brief
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 2026-10-07
Covers 2026-09-24 → 2026-10-07. Anchored on OpenAI DevDay (September 29). Picks drawn from Google's move from Gems to skills, Gemini Notebook grounding arriving in Docs, Gemini Enterprise querying data in place, DevDay's push into long-running agents, and Claude Sonnet 5.5 lowering the everyday cost of the FY27 Claude plan. Hand-curated this fortnight; the scheduled filter wires up before the next digest.
Skills replace Gems across Gemini and Workspace — existing Gems retire from March 2027
The news. Google introduced skills — reusable, Markdown-based instruction sets invoked with "/" in chat — that work across both the Gemini app and Workspace apps, can be stacked together, and can carry reference files. Workspace rollout began October 5 for Rapid Release domains (October 19 for Scheduled Release); Gems move into Settings on November 17 and lose create, edit, and use on business and enterprise editions from March 1, 2027. Existing Gems migrate automatically, but as inactive drafts that only their creators can switch on, with Drive sharing carried over.
Why this matters for Scotts. Every Shared Gem a team depends on — and every Workspace Studio flow with an "Ask a Gem" step — needs its owner to activate or rebuild it before March, so the Gemini Tools & Ecosystem Guide should start teaching skills now and Champions should inventory the Gems their teams rely on.
Docs can now ground Gemini answers in your Gemini Notebook sources — with inline citations
The news. Typing "@" in the Gemini side panel or bottom bar in Google Docs now lets users reference an existing Gemini Notebook, so drafting and Q&A in Docs draw on that notebook's sources with inline citations instead of a single file or the open web. It is live on both Rapid and Scheduled Release for Business Standard/Plus, Enterprise Standard/Plus, and Education Plus, and requires Gemini for Workspace in Drive (on by default where already configured).
Why this matters for Scotts. This is the missing last step for the Auto-load weekly reports into NotebookLM idea — once the weekly PDFs land in a notebook automatically, the weekly summary can be written in Docs grounded in those same reports, which makes the full pipeline worth demoing at an upcoming office hours.
Gemini Enterprise can now query BigQuery, Cloud SQL, AlloyDB, and Spanner in place — no ingestion required
The news. A new federated query mode (Preview) lets Gemini Enterprise query AlloyDB for PostgreSQL, BigQuery, Cloud SQL, and Spanner directly through the Model Context Protocol, using the signed-in user's own credentials instead of copying data into a separate data store. Knowledge Catalog integration also entered Preview, and granular IAM permissions for Gemini Enterprise apps and data stores went GA on September 28.
Why this matters for Scotts. The Production Planning Tool works today by pulling fragmented data into one place with generated code; wherever that data already sits in BigQuery, in-place querying under each user's own permissions is a simpler, better-governed path, and is worth a Preview test against one existing dataset before the pilot scales.
Next two: outside the Gemini stack — OpenAI DevDay pushes long-running agents into the enterprise, and Claude Sonnet 5.5 lowers the everyday cost of the FY27 Claude plan.
OpenAI DevDay: Dots agents keep working after the hand-off, and a new Agents API targets multi-agent builds
The news. OpenAI introduced Dots, agents that keep working on a task after the initial instruction, operating across ChatGPT, email, Slack, and text messages with their own cloud compute and browser — initially for Pro, Business Premium, and Enterprise users. Alongside it came GPT-6.1 Sol (near-flagship performance at lower cost, with a 1.05M-token context window), ChatGPT Spaces for team collaboration, Codex Cloud, and an Agents API with computer use and multi-agent orchestration.
Why this matters for Scotts. Long-running, cross-app agents are now the shared direction across Google, Anthropic, and OpenAI, which gives the FY27 platform evaluation a clear benchmark for Bloom — the questions worth pressing are governance and where agent output lands, since Bloom is designed to put it back into Workspace.
Claude Sonnet 5.5: the everyday tier gets 30% faster and up to 30% cheaper per task
The news. Anthropic released Claude Sonnet 5.5, which nearly matches Opus 5.5 on knowledge-work benchmarks, generates output more than 30% faster than Sonnet 5, and uses fewer tokens per task — cutting typical costs by up to 30% at unchanged pricing of $2 / $10 per million input/output tokens. It is available in the Claude apps, the Claude API, AWS, Google Cloud, and Azure, and is the first Sonnet model with cybersecurity safeguards comparable to Opus 5.5.
Why this matters for Scotts. Sonnet is the tier most associates would use day to day under Claude Enterprise, so the FY27 cost model in the Governance Committee's platform brief should be refreshed with Sonnet 5.5's per-task savings, not just its unchanged per-token price.
Top 5 — week of 2026-09-23
Covers 2026-09-10 → 2026-09-23. No major anchor event this fortnight — picks drawn from Gemini Notebook's biggest source expansion yet, Workspace Studio's automation leap, and two Claude Enterprise updates that sharpen the FY27 cost and governance case. Hand-curated this fortnight; the scheduled filter wires up before the next digest.
Gemini Notebook can now draw on 100,000+ published books as grounded sources
The news. Google launched Expert Intelligence in Gemini Notebook on September 17 — a cross-Google initiative that lets users add eligible Google Play Books ebooks directly to a notebook as a citable source. The catalogue covers over 100,000 titles from Penguin Random House, Macmillan, O'Reilly Media, Bloomsbury, Johns Hopkins University Press, and others. Once added, Gemini Notebook generates answers grounded specifically in the book's text alongside any other uploaded sources. Audio Overviews, Infographics, and Quizzes can be generated from book content. Users must own the ebook through Google Play Books; if a notebook is shared, collaborators are prompted to purchase their own copy. Admins must have Gemini Notebook, Google Play, and Google Books enabled for the organisational unit.
Why this matters for Scotts. The NotebookLM Enterprise intake card positions Gemini Notebook as a knowledge management tool — Expert Intelligence significantly extends that value proposition by grounding research in authoritative published sources, not just internal documents. For teams involved in strategy, operations research, or professional development, the ability to combine internal context (planning docs, meeting notes) with expert literature in a single grounded session is a meaningful capability upgrade worth including in any Gemini Notebook pilot proposal.
Workspace Studio flows can now trigger from external apps, call Slack and Salesforce, and fire webhooks
The news. Google added four capabilities to Workspace Studio flows from September 17: custom starters (trigger a flow from events in other apps), custom steps (run Google Apps Script logic inside a flow), third-party integrations in beta (read from and write to Asana, Confluence, HubSpot, Jira, Mailchimp, QuickBooks, Salesforce, and Slack), and webhooks (send HTTP requests to external endpoints). Admin console settings roll out September 17; end-user features follow September 21 for Rapid Release domains and September 30 for Scheduled Release. All features sit behind granular admin controls. Webhook URL allowlisting is available on Business Plus, Enterprise Standard, and Enterprise Plus.
Why this matters for Scotts. Workspace Studio previously operated only within Workspace apps — this update opens it to the broader toolset Scotts teams actually use. A flow can now start when a Jira ticket changes status, pull context from Salesforce or ServiceNow, run a custom Apps Script step, and send results to a Slack channel — all without IT writing a custom integration. This directly enables the kind of cross-system automation that the Agentic Workflow Builder intake card has been scoping, and deserves a proof-of-concept conversation with at least one operations team before year end.
Gemini can now create Docs, Sheets, and Slides for you from anywhere in Workspace — without switching tabs
The news. Gemini's cross-app orchestration capability reached general rollout from September 9. Associates can now prompt Gemini from Gmail, Drive, Docs, Slides, or Chat to create content in another Workspace app — a formatted Doc, a structured Sheet, or a styled Slides deck — without leaving the app they are working in. Gemini acts as an intelligent orchestrator, pulling real-time context from relevant files, emails, and chat threads to personalise the output. Usage of advanced AI features is subject to plan-level usage limits.
Why this matters for Scotts. The most common friction point in Workspace AI adoption is context switching — associates interrupt their flow to open a different app, rebuild context, and prompt from scratch. Cross-app orchestration removes that step: a supply chain planner reviewing an email can ask Gemini to generate a summary Slide or a structured Sheet without leaving Gmail. This is the single most visible daily-use improvement since Gemini launched in Workspace, and is worth highlighting in the next Office Hours session as a concrete, try-it-today capability for all associates on Business Standard and above.
Next two: outside the Gemini stack — Claude Opus 5.5 resets the FY27 cost model again, and Claude Enterprise Smart Reports give admins the usage intelligence to build the adoption case.
Claude Opus 5.5 launches — Fable 5.1 performance at 40% lower cost
The news. Anthropic launched Claude Opus 5.5 on September 22 — the first model in the new 5.5 family. It performs at the level of Claude Fable 5.1 on most workloads and costs 40% less to run than Opus 5. It carries a 1M token context window, 128K maximum output tokens, and always-on adaptive thinking. The model ID is claude-opus-5-5. Anthropic positions Opus 5.5 as the recommended starting point for long-running agentic coding and knowledge work, with Sonnet 5 serving lower-cost agent workflows and Haiku 4.5 as the lightweight option.
Why this matters for Scotts. Opus 5.5 effectively collapses the cost gap between the Opus tier and the Fable tier — for most complex agentic workloads, Scotts can now get near-Fable performance at Opus pricing. The FY27 Claude Enterprise cost model should be updated: Opus 5.5 at ~$3.50/$17.50 per million tokens (estimated at 40% below Opus 5's $5/$25) now sits as the practical workhorse tier, with Fable 5.1 reserved only for the most demanding document-heavy or reasoning-intensive workflows. Worth refreshing the platform evaluation brief before the FY27 budget cycle closes.
Claude Enterprise Smart Reports give admins visibility into team usage, cost, and adoption friction
The news. Anthropic launched Smart Reports in beta for Claude Enterprise — automated analysis of how a team is using Claude, reporting on work completed, cost per session, where sessions hit friction, and which repeated patterns are worth packaging as shared skills. Enterprise admins receive these reports; Pro and Max individual plan users get a version first, with Enterprise admins receiving at least 30 days notice before any changes to their organisation's reporting.
Why this matters for Scotts. The FY27 Claude Enterprise evaluation has lacked one key input: evidence of what value adoption actually delivers and where it stalls. Smart Reports provide exactly that — a usage-intelligence layer that shows which teams are getting value, what it costs, and where friction is preventing adoption. For the AI Governance Committee, this is also a built-in compliance and oversight tool: usage patterns, cost attribution, and repeated workflows are all visible without requiring custom analytics. Worth requesting early access as part of the FY27 platform trial.
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.