Research · 2026-04-18 · source issue

Balsam AI Landscape — Discovery Pass

A one-time audience-organised scan of AI work visible from Balsam's Atlassian properties (Jira + Confluence), with an explicit gap analysis of under-served functions.

A one-time discovery scan of AI work visible from Atlassian (Jira + Confluence). Audience-organised so a Balsam employee can find themselves and the tools that match their work. Gap analysis is the most valuable part — the audiences with no AI footprint yet are where a cohort could partner.

Scope limits: Atlassian only. Anything lived in SharePoint, Slack, OneDrive, email, or private GPT libraries is invisible to this scan. That’s called out again in Methodology.

Summary

What exists. Balsam has a governance spine (an AI Council charter, a legal AI-use notification framework, an MCP Server Registry), a heavily-developed developer surface (Claude Code + Cursor bake-off, agent catalogue, CLAUDE.md standards), and a few mature non-engineering practices (SEO Custom GPTs, a Product Management 30-60-90 Claude adoption plan, a live ChatGPT-via-MCP analytics app in production, a set of Agile PM Claude prompt playbooks). The 41-ticket Jira AI project is an ~8-week-old, engineering-centred body of work.

What’s growing. A self-serve analytics stack — Snowflake + Cortex Analyst + Ask Sigma + dbt semantic layer — is the biggest non-engineering-facing bet. It’s already in production for ChatGPT, with Claude MCP being wired up alongside. Agentic-commerce (Google UCP, OpenAI ACP, PayPal agentic via Braintree) is moving out of Buyflow into a Marketing/Feedonomics workstream. SEO is running the most mature Custom GPT roster in the company.

What’s missing. Twelve of the fourteen largest functional teams (Marketing, Creative, Customer Service, Finance, HR, Retail, Operations, eCommerce Ops, Performance Marketing, Engagement Marketing, Merchandising, Indirect Sales) have zero AI footprint on Atlassian. Some of this is genuine (no AI adopted) and some is displacement (the work lives in SharePoint, email, or private ChatGPT accounts). Either interpretation is a CoE opportunity — a cohort built around one of these teams would have an open field.

Gap analysis — audiences under-served

These functions appear in Balsam’s org chart but have no AI-related Confluence pages authored by their team and no AI-related Jira tickets assigned to their team. For each, we list what they already use today (an AI-consumable surface) and where an invitation could begin.

Marketing (Brand, Performance Marketing, Engagement Marketing, Print/Direct Mail)

• Footprint: None. No AI pages authored in the MAR, PM, or EM spaces.
• Existing systems the team already uses: ClickUp (primary creative-request tracker — “Creative Deliverables” list across Print, PPC, Paid Social, Display), Optimove (status tracker tracked weekly), Canva Magic Studio (listed in the Oct 2024 GenAI tools matrix as “in beta, in use” for Marketing, but no process documentation), Ziflow (Creative + Marketing), SharePoint for campaign briefs, Smartsheets for direct-mail tracker.
• What’s agent-ready (data the team already generates): The ClickUp creative-requests list, BHI Commercial Briefs in SharePoint, the Direct Mail Tracker Smartsheet, monthly Promo Calendar in SharePoint. None of this is wired to an AI agent today.
• Invitational framing: A cohort that takes a Marketing Ops question — “where is campaign X in its lifecycle?” — and answers it by reading ClickUp + SharePoint briefs + the promo calendar.

Creative

• Footprint: None in the CREAT space.
• Existing systems: Adobe Creative Cloud (Firefly, Generative Fill — listed in the GenAI tools matrix but no process doc), Figma, Ziflow, ClickUp (intake).
• What’s agent-ready: Widen DAM asset catalogue, Figma component libraries, ClickUp brief structure.
• Note on the Figma MCP Integration Assessment (Digital Engineering, Oct 2025): an engineering-side evaluation of Figma MCP exists — Creative has no authored counterpart.

Customer Service

• Footprint: None in the CS / TCP spaces.
• Existing systems: Gladly (with Gladly Sidekick “conversational chatbot to handle common requests… order statuses and handle return and cancellation requests” — listed as “in use” for Season 2024).
• What’s agent-ready: Gladly Sidekick is already the most production-deployed AI feature at Balsam and has no documentation or performance-measurement page in Confluence.

Finance & Accounting

• Footprint: None in the FA space. AI-adjacent work lives in the Shared ERP Documentation (SED) space — but it’s about Finance workflows, not authored by Finance.
• Existing systems: NetSuite (no native Oracle AI features documented), Stampli (AP automation, “simplifies Balsam Brands’ AP complexity across global, seasonal entities” — vendor proposal on file, no implementation page), AWS Cost Explorer (engineering-facing CustomGPT shipped; Finance-facing equivalent in backlog as AI-23 Azure Cost Explorer).
• What’s agent-ready: NetSuite data, vendor-management documentation, Stampli invoice data once implemented.
• Invitational framing: A Finance cohort could build the NetSuite user-interview CustomGPT that’s already proposed in a PjM brainstorm (“Job Tasks and Activities Discovery Bot”) — or formalise the AWS Cost Explorer GPT pattern for Finance consumption.

People / HR

• Footprint: None. The only People-space page touching AI is an onboarding guide template.
• Existing systems: HR platforms (not surfaced in Confluence), Organimi (org chart).

Operations & Supply Chain

• Footprint: Minimal. The OPS space has no authored AI material; OPS-3802 (IBM OMS security remediation) consumed a Claude-authored security report as an input.
• Existing systems: IBM OMS, warehouse integrations (WIS/OMS/IIT/Snowflake — the ET space has a 2026 Overview page), Project Phoenix ERP/NetSuite work.

Retail / Wholesale (BH Studio, Indirect Sales, GE/Lowe’s)

• Footprint: One Indirect Sales page mentions an “AI-assisted PO match checklist” for the GE Lowe’s direct-import flow. That’s the entire user-side AI story for this function.
• Existing systems: Squarespace (for GE Holiday Lighting microsite eval), SAP Commerce / Hybris (for wholesale B2B).

eCommerce Merchandising & Merch Ops

• Footprint: None in the EC space. The BOW-17329 ticket (“Review sample names per product type so Merch Ops’ ChatGPT-based validator can check child/parent/family names”) suggests a Merch-Ops-authored ChatGPT exists but is invisible to Atlassian.
• Existing systems: Riversand/Contentserv PIM, Feedonomics product feeds, Hybris, Constructor (site search).
• What’s agent-ready: The product feed (AI-42 Create Product Feed Agent in backlog already names this), PIM exports, category taxonomy (flagged in Category-Based Promotions — PRD as consistent-across-Hybris-Impact-Analytics-Constructor).

Digital Tools Optimization Pod (CMS & Localisation workflow)

• Footprint: Heavy PDM/Project Management activity on future-state workflow design, much of it Claude-drafted and Claude-reviewed. The team has already absorbed “Uncle Claude” into casual vocabulary — a telling signal of embedded use.
• Tension worth naming: A comment thread on the PRD flags that marketing → creative → publish runs on ClickUp, tech runs on Jira, and there is no Jira↔ClickUp integration. Whether that’s an integration problem or an agent-orchestration problem is an open question.

Executive / Leadership

• Footprint: None authored for the executive persona.
• What exists adjacent: [Product Team Weekly Update — Week of March 23, 2026](https://balsambrands.atlassian.net/wiki/spaces/PDM/pages/1271202305/Product+Team+Weekly+Update+Week+of+March+23+2026) was “Created by Claude | Sources: Confluence Phoenix Forge recordings, Jira (Blocked tickets), Outlook Calendar” — that pattern could serve other leadership surfaces.

By audience

Engineering & Data Engineering

The most mature and best-documented audience. Three layers:

Governance / standards (Digital Engineering space).

• AI Council Charter — five pillars (Policy, Ethics, Security, Agentic Governance, Education). Monthly full council + quarterly C-suite scorecard.
• AI Council Policies and AI Use Case Intake & Request Process — the org’s approval gate.
• MCP Server Registry — “authoritative registry of all approved custom MCP servers… Any MCP server not listed here is unauthorized.” Owned by Enterprise Architecture.
• Custom MCP Server Governance & Standards — operating standards.
• Legal: AI Use Notification & Approval Framework (Legal space) — three-level framework with an AI Use Registry for external-facing use.

IDE / tooling.

• Claude Code + Cursor bake-off (AI Coding Tools - Final Comparison, plus Cursor/Claude Code per-tool development-experience docs). Current posture: “Leads & Managers: Use both. Developers: Use Cursor as primary tool but with Claude.”
• How to Set Up Atlassian MCP with Claude Code (Local) — Feb 24, 2026 runbook.
• Figma MCP Integration Assessment — design-to-code eval.
• Claude Training Recordings — TA + TM/TI session recordings and a Balsam_Vibe_Workshop_Session_Deck.
• Engineering plug-in platform: Jira Epic AI-31 “AI Engineering,” including AI-30 “BVH Plugin Creator” — a meta-plugin /bvh:create-plugin that generates installable Balsam Vibe Hub agent-team plugins via a 7-phase wizard.

Practice (the AI space and Jira AI project).

• The AI Confluence space is developer-focused: Starting a Project with Claude Code, How to Write a CLAUDE.md, Claude Code Concepts, How to Work With Agents, Available Agents catalogue.
• Available Agents catalogue (ready-to-copy .md files): service-architect, refactor-engineer, code-debugger, technical-analyst-writer, qa-test-fixer.
• AI Projects and Initiatives — currently an empty template awaiting population.
• Jira AI project (41 tickets, Feb–Apr 2026): a Claude POC debugging WMS order API (AI-26), a support-engineer agent team (AI-28), a multi-agent security swarm (AI-32), Bitbucket code-review integration (AI-35), an Application Manager for engineering projects (AI-41).
• Claude Code training sprint run inside the Buyflow pod (17 BUY-* training subtasks).

AI-consumable surfaces: Bitbucket, Jira, Confluence, Atlassian MCP, the Balsam Vibe Hub plugin manifold.

Data & Analytics Engineering

The second most mature audience, and the only one with a business-user-facing AI app in production.

• Live in PROD: ChatGPT access to Snowflake PROD via a Custom GPT and a Snowflake MCP server, plus a Streamlit self-serve app (Self-Serve Analytics — Master Index). “Status: Active — ChatGPT MCP and Streamlit live in PROD.”
• Next up: Claude-via-MCP as a second front-end (DAE-370, DAE-252, DAE-295/296 governance prep); Cortex Analyst as the shared semantic API; Ask Sigma Phase 1 (DF-3990 done) → Phase 2 (DAE-421).
• Infrastructure: Containerised Snowflake MCP server on EKS (DF-4078), OAuth + RBAC for Claude Co-work and for ChatGPT, a dbt-backed semantic layer, an Airbyte/Snowflake/New Relic daily agent (DAE-369) posting errors to #data-engineering-alerts, Cortex-based cost monitoring (DAE-339, done), a Claude Desktop plugin balsam-airbyte-eks-log-analyzer (DAE-282, done).
• Claude-in-dev-workflow adoption: DAE-276 “Automated PR & SQL Reviews,” DAE-264/DAE-303 dbt-docs/Claude integration, DAE-337 Claude-generated unit tests + docs as a CI gate.
• Data catalog prerequisite: DF-3394 (select data catalog tool for dbt — blocked).

AI-consumable surfaces: Snowflake (read-only, via a restricted role), Cortex Analyst semantic API, Ask Sigma, dbt catalog, New Relic, Airbyte logs.

Product Management

The most mature non-engineering audience.

• PM Team Claude Adoption Plan (30-60-90 Days) — “Goal: Replace ChatGPT for daily writing tasks — low friction, immediate ROI. Stakeholder comms: Use Claude to draft status updates, exec summaries.”
• AI Revenue Estimation System — auto-scores Buy/Shop-pod Jira tickets with Low/Mid/High value + confidence tier + dev-complexity sizing when triggered on a Story/Bug.
• AI Working Session: Building Jira Ticket Templates — a cross-functional prompt-engineering workshop.
• Product Team Weekly Update — Week of March 23, 2026 — “Created by Claude | Sources: Confluence Phoenix Forge recordings, Jira (Blocked tickets), Outlook Calendar.”
• Agentic Checkout via LLMs: Options, Considerations, and Decision Framework — evaluation for 2026 peak season.
• Shared Claude Project with reference files for Digital PdM (AI-39) seeded with contextual docs for Shop/Buy/Mkt pods.
• User Story Skill (AI-36) and Revenue Estimator for Jira Tickets (AI-37) skills shared with digital pods.
• CMS & Localisation future-state work (CMS & Localization Future State Workflow v2, [4/1 Review of Future State Doc](https://balsambrands.atlassian.net/wiki/spaces/PDM/pages/1283325987/4+1+Review+of+Future+State+Doc) — “Using Claude to help review the initial draft… The Future State is a well-structured AI-generated document…”).

AI-consumable surfaces: Jira (Stories/Bugs + custom fields for estimates), Confluence, Phoenix Forge Zoom recordings, Outlook Calendar, product category taxonomy.

Agile Project Management (PjM, MKT pod, scrum masters)

• Claude Sprint Capacity Report — SP Rules & Process Reference — “This is Claude’s persistent memory for the MA Sprint Capacity Report. Read this page at the start of every new conversation before running the report. Just say ‘run the capacity report.’”
• Claude Prompt Playbook series (Apr 2026): Bug Rate Analysis, Rework Rate Analysis, Scope Creep Analysis, MKT Release Health Check, MKT Blockers Matrix.
• MKT Sprint & Release Health — “AI-generated reports that pull live data from Slack and Jira, produce an interactive HTML file.”
• Ticket Intake — AI Prompts for Product, MA Initiative Creation — AI Prompt for Product.
• Jira AI-9/AI-10/AI-11/AI-12/AI-13/AI-14 — an APM 2026 roadmap story set (capacity planning, dependency/blocker detection, sprint status reports, retrospective insights, backlog refinement, scope-creep detection).
• Jira AI-44 (Done): Claude prompt generates Initiative → Epic → Story/Task/Spike hierarchy from a business-case source.

Named CustomGPT proposal worth surfacing to a non-engineering cohort: a PjM brainstorm, “Brainstorm: Job Tasks and Activities Discovery Bot” — “Could we create a GPT that we could send to users of NetSuite or users impacted by the PD2L project that would interview them to capture their specific job tasks and activities?”

AI-consumable surfaces: Jira (capacity custom fields per discipline, used by the capacity-report workflow), #scrum-of-scrums, #pod_marketing, #pod_leads_marketing Slack channels, Confluence, Loom recordings.

Quality Engineering

• AI-43 Automated failed-test analysis + Confluence reporting (in progress): Claude in CI/CD collects post-retry failures, generates a structured failure report, overwrites a GEO-specific Confluence page each run to distribute ownership.
• AI-3 Done: CustomGPT pulls Zephyr Scale test case IDs from CSV and publishes automation reports to Confluence.
• AI-27 CustomGPT for Bitbucket integration (lists/filters PRs, retrieves diffs, triggers pipelines).
• AI-25 Skills.md for Codex so generated Playwright tests conform to codebase protocols.
• AI-26 Claude POC debugging WMS order API.

AI-consumable surfaces: Zephyr Scale, Bitbucket, CI/CD logs, Confluence.

DevOps

• AI-35 AI-powered code reviews wired into Bitbucket Pipelines (in progress).
• AI-32 Multi-agent security swarm for vulnerability testing (in progress).
• AI-28 Slack-integrated Support Engineer Agent Team reading user-reported issues, pulling log context, posting cause/solution hypotheses to an engineer review channel.
• AI-1 Claude for Production Support (spike, backlog).
• New Relic Grok for site monitoring — listed in the GenAI tools matrix as DevOps testing.
• Infra: DSO-4847 POC AI code-review agent team for Bitbucket, DSO-4804 token rotation for ai-assisted-code-reviews and claude-review Bitbucket bots.

AI-consumable surfaces: Bitbucket, New Relic, CloudWatch logs, AWS Secrets Manager, Slack (ops channels).

SEO

The non-engineering function with the most mature AI practice.

• AI Exploration — a named Custom GPT roster:
  • SEO Analyst — compile recurring reports, pulling the data and compiling the initial report.
  • LOC Agents — localise content (articles, SEO text blocks) from US-ENG to geo-specific ENG (UK / AU / CA-ENG).
  • PLP FAQ — generate unique FAQ questions for the SEO text blocks on PLPs.
  • Schema Generator / Review — generate or review schema.
• 2026 MNL Summit Debrief — “SEO is evolving from a rankings-driven channel to a discoverability and influence engine… Discovery now happens across: Search (Google, AI Overviews / SGE), AI platforms (ChatGPT, Perplexity), Social (TikTok, Instagram), Community (Reddit, reviews).”
• [LLMS TXT Script](https://balsambrands.atlassian.net/wiki/spaces/SEO/pages/1210548239/LLMS+TXT+Script), [LLMS-PRODUCTS TXT Script](https://balsambrands.atlassian.net/wiki/spaces/SEO/pages/1216643094/LLMS-PRODUCTS+TXT+Script), [Process - Creating an LLM Text File](https://balsambrands.atlassian.net/wiki/spaces/SEO/pages/1211367440/Process+-+Creating+an+LLM+Text+File) — a Python pipeline that reads the sitemap and emits llms.txt for LLM crawler consumption.
• [Ai Performance Measurement](https://balsambrands.atlassian.net/wiki/spaces/SEO/pages/1182040097/Ai+Performance+Measurement) — New Relic monitoring of ChatGPT, Perplexity, and Google-Agent crawler traffic.
• SEO-3654, SEO-3652 — SEO GPT non-English slug + localisation writing agents (QA in progress).

AI-consumable surfaces: sitemap + llms.txt, Sitecore/Contentstack-equivalent CMS content, Wikidata (SEO-3680 spike).

ERP / Project Phoenix (Finance-adjacent, not Finance-authored)

• [Claude Conversations - Source](https://balsambrands.atlassian.net/wiki/spaces/SED/pages/1253605380/Claude+Conversations+-+Source) — full Claude conversation log published to Confluence from Project Phoenix gap-analysis work (Director, Omnichannel Product Management + Claude).
• Project Phoenix — Documentation Gap Registry — a systematic Claude-driven read of the Project Phoenix Confluence.
• [Wholesale Order Flow — NetSuite One World (BBI/BUIC FOB Model)](https://balsambrands.atlassian.net/wiki/spaces/SED/pages/1229848577/Wholesale+Order+Flow+NetSuite+One+World+BBI+BUIC+FOB+Model) — “Created via Claude — March 2026.”
• Purchasing and Allocations BRDs written in a format tuned for LLM ingestion.
• Meeting Notes – Vendor Management Process Alignment & NetSuite Design — Loom AI recording + Read.ai report in active use among NetSuite / ERP stakeholders.
• Stampli AP-automation follow-up proposal on file (no implementation page yet).

Wholesale / Indirect Sales

• [GE Direct Import — Lowe's](https://balsambrands.atlassian.net/wiki/spaces/INDR/pages/1201602572/GE+Direct+Import+Lowe+s) — references an “AI-assisted PO match checklist (Balsam_PO_Match_Checklist.xl…).” The only user-authored AI footprint in Indirect Sales.

Enterprise Technology

• [WMS Developer Onboarding](https://balsambrands.atlassian.net/wiki/spaces/ET/pages/1214808115/WMS+Developer+Onboarding) — “Automated (recommended) — Use Claude Code with the /wms:setup command for a fully guided setup.”
• [COS User Role Permissions Matrix](https://balsambrands.atlassian.net/wiki/spaces/ET/pages/1247019232/COS+User+Role+Permissions+Matrix) — “Created using Claude, needs human confirmation.”
• [Warehouse Integrations 2026 Overview](https://balsambrands.atlassian.net/wiki/spaces/ET/pages/1283981314/Warehouse+Integrations+2026+Overview) — WIS/OMS/IIT/Snowflake status board (agent-ready data surface).

Strategy / Leadership / Cross-functional

• The GenAI Council (from the older 2023–2024 GenAI space) is the predecessor body to the AI Council. Membership listed in the GenAI space was: COO + Emma Keane + Joe Balczo + Kiran Ramesh + Setu Garg + Vince Tiu + Caroline Tuan (Exec sponsor). Treat as historical context for the AI Council’s mandate.

Data sources — the AI-consumable landscape

Systems Balsam teams already use and could point an agent at. Listed by ownership so a persona question (“how would a Finance person…”) can find the matching data.

MCP servers & integrations

All MCP work runs through the MCP Server Registry (“any MCP server not listed here is unauthorized”). The visible footprint:

• Snowflake MCP — Managed connector with OAuth + RBAC, exposing a narrow, governance-reviewed set of fact tables to a read-only role. Containerised on EKS. Operational specifics (account, role names, PAT lifecycle, exposed tables) live in the setup runbook inside Confluence.
• Atlassian MCP — In use from Digital Engineering (including from this repo). Runbook for Claude Code local setup published Feb 24, 2026.
• Figma MCP — Engineering-side integration assessment done Oct 2025 (design-to-code pipeline).
• ChatGPT Snowflake MCP — Live in PROD for self-serve analytics, separate from Claude MCP.
• Claude Desktop plugin: balsam-airbyte-eks-log-analyzer — Data Engineering, Airbyte EKS sync diagnostics.
• Bitbucket MCP / CustomGPT — QE CustomGPT for PR-list / diff / pipeline / build-info (AI-27).

Not yet MCP-wired but frequently named as candidates: ClickUp (no trace), NetSuite, Gladly, Widen, Feedonomics, Constructor, SharePoint, Optimove.

Models & platforms in use

• Claude (Anthropic) — The default for Digital Engineering, Data Engineering, Product Management, Agile PM, Project Phoenix / ERP, WMS, Enterprise Technology, CMS / Localisation work.
• ChatGPT (OpenAI) + Custom GPTs — Default for SEO (named roster), Self-Serve Analytics (live Custom GPT), some Product Management daily writing, Merchandising-Ops validation (BOW-17329), Consumer Insights scoping (AI-2), Zoom integration exploration (AI-29).
• Rovo / Rovo Agents (Atlassian) — Named in the APM 2026 roadmap story set; no other adoption surfaced.
• GitHub Copilot — Engineering, per the 2024 GenAI tools matrix and the AI-Agnostic Coding Standards doc.
• Cursor — Engineering; tied with Claude Code in the formal bake-off, preferred for developers.
• Microsoft 365 Copilot — “Testing — users have not been impressed” (2024 matrix); no evidence of growth since.
• Perplexity — Named in SEO crawler monitoring and discoverability strategy; not used for internal work.
• Gladly Sidekick — Customer Service conversational chatbot, “in use” for order status / returns / cancellations.
• New Relic Grok — DevOps, testing.
• Read.ai, Loom AI, Zoom meeting summaries — Meeting productivity, ad-hoc.

Backlog & intentions

Non-exhaustive; the freshest public signals of where AI is heading:

• Azure Cost Explorer CustomGPT (AI-23, backlog) — Finance-facing parity with the existing AWS Cost Explorer GPT.
• Product Feed Agent (AI-42, backlog) — custom app so developers can consume the product feed locally.
• CMS agents for eCommerce Ops (AI-5, backlog) — no owner yet.
• Audit Agents for ERP/OMS anomaly scanning (AI-4, backlog) — no owner yet.
• Multi-agent assistant for engineers (AI-22, backlog) — intake / analysis / planning / engineering / review of tickets via Claude API.
• NetSuite user-interview CustomGPT (PjM brainstorm: Job Tasks and Activities Discovery Bot) — proposed for the PD2L project; nobody has picked it up.
• Agentic-commerce production launch (BUY-2836 Google UCP + OpenAI ACP, BUY-3092 PayPal via Braintree, MA-2321 Feedonomics agentic feeds) — US-only at launch, 2026 peak season.
• Category-aligned taxonomy across Hybris / Impact Analytics / Constructor — prerequisite for agentic-commerce feeds and for AI-readable merch data.
• Data catalog tool selection for dbt (DF-3394, blocked) — prerequisite for the semantic layer’s public-facing docs.
• Ask Sigma Phase 2 (DAE-421) — Cortex Agent.
• APM 2026 roadmap (AI-9 through AI-14) — AI-driven sprint reports, dependency detection, retrospective insights, backlog refinement, scope-creep detection, capacity planning.

Methodology

Queries run against cloudId d7df471e-35bf-4490-a4ca-00bdafb42695 on 2026-04-18. Re-running these queries is the baseline for the daily regeneration in #25.

Jira (JQL):

project = AI ORDER BY updated DESC

project != AI AND (text ~ "LLM" OR text ~ "Claude" OR text ~ "ChatGPT" OR text ~ "CustomGPT")
  AND updated > -540d ORDER BY updated DESC

project != AI AND (text ~ "MCP" OR text ~ "Copilot" OR text ~ "Cursor" OR text ~ "\"Model Context Protocol\"")
  AND updated > -540d ORDER BY updated DESC

text ~ "ClickUp" AND updated > -540d ORDER BY updated DESC

Confluence (CQL):

space = AI AND type = page ORDER BY lastmodified DESC

space = GenAI AND type = page ORDER BY lastmodified DESC

(text ~ "Claude" OR text ~ "ChatGPT" OR text ~ "CustomGPT" OR text ~ "LLM")
  AND lastmodified > "2024-10-01" AND space != AI AND space != GenAI ORDER BY lastmodified DESC

(text ~ "MCP" OR text ~ "\"Model Context Protocol\"" OR text ~ "Copilot" OR text ~ "Cursor"
  OR text ~ "\"prompt engineering\"")
  AND lastmodified > "2024-10-01" AND space != AI AND space != GenAI ORDER BY lastmodified DESC

text ~ "ClickUp" AND lastmodified > "2024-01-01" ORDER BY lastmodified DESC

Directly fetched pages (full-body read):

• 1162313748 — AI Projects and Initiatives (AI space) — empty template.
• 1163461090 — AI space overview.
• 17301528 — Tools with embedded GenAI features (GenAI space, Oct 2024).
• 1198293003 — AI Council Charter.

Spaces surveyed: all Confluence spaces returned in the first getConfluenceSpaces page (~100 spaces, all that existed). Personal spaces are included in CQL hits.

Known blind spots.

• Non-Atlassian sources are invisible. No Slack, SharePoint, OneDrive, Google Drive, Notion, email, or private CustomGPT libraries were scanned. Several strong signals point to AI work living outside Atlassian (BOW-17329 implies a Merch Ops ChatGPT validator that has no Confluence page; Marketing uses SharePoint-primary briefs; Finance teams’ correspondence with Stampli is over email/PDF). A follow-on phase should at minimum ask each function whether they have a CustomGPT or Claude Project that isn’t registered with the AI Council.
• ClickUp, the Marketing/Creative workhorse, has no AI surface. It is referenced in 50+ Jira tickets and ~60 Confluence pages as the request-tracking tool of record, but there is no MCP server, no ChatGPT / Claude integration, and no Atlassian-visible automation. A Marketing cohort could make this the highest-leverage single integration in the company.
• Atlassian MCP scope. This scan used the Atlassian MCP connection from this repo (alex.finnemore@balsambrands.com). If other team members have access to spaces this account does not, those would also be blind spots.
• Name attribution. Names found in Jira assignees and Confluence authors were checked against a Balsam org-chart snapshot. Pod/team attribution was used throughout; the directory search surfaced variants (e.g. “Gabi Viggiano” vs “Gabriela Arriaga-Viggiano”; “Charles De Long” vs “Charles DeLong”; “Rich Nicasio” vs “Richmond Mark Nicasio”) and a handful of names not present in the directory (which may be contractors or recent joiners) — kept out of this document where possible, and flagged for verification on #25 regeneration.
• “MCP” / “Copilot” / “Cursor” are ambiguous tokens. False positives stripped from the cross-project sweep: “MPC” = Multi Payment Checkout; “Cursor” = mouse cursor in many PjM whiteboards; “Copilot” was essentially unused outside Microsoft 365 Copilot mentions. Retained count: ~50 distinct initiatives across 100+ raw hits.