Pages
Users / Devices v1
Source: bob-wire c21.inventory — Inventory + Users/Devices split, from the 2026-08-31 Mark Peterson advisor call
The people view of the c21 split: who is using AI, on what devices, with what accounts — and where the sensor ISN'T deployed. Machines moved here from Inventory; agent attributes appear only as filters. The roster is the teams-ideas-v7 idiom on the layer DataTable (#details expand, column search), the coverage header answers Mark P's #1 story — "what's the deployment status of Qpoint sensor across my user base?" — against the Okta/Jamf denominator, not the sensor's own sightings.
Users / DevicesWho's using AI, on what devices, with what accounts — and where the sensor isn't. name TBD — "Teams & Personnel" also floated
Sensor is on 12 of 18 known devices — 3 people have no coverage at all
Mon 17:00 UTC✓ Okta connected✓ Jamf connected15 people / 18 devices known denominator = identity/MDM source of truth, not what the sensor happens to see
12/18devices sensored
6 without — 3 of those are the Windows gap
67%device coverage
vs the Okta/Jamf denominator
12/15people covered
3 with no sensored device
1investigate
Alexis — seat, zero telemetry
Not covered — actionable, not a footnote"who has the sensor installed, and who doesn't — and that should be actionable"
AR
Alexis RomeroEngineering · alexis-mbp · no sensorholds an active enterprise Anthropic seat — zero telemetryMT
Miguel TorresGTM · miguel-mbp · known to Jamfno sensor deployed — no AI usage observedOH
Omar HaddadOps · omar-mbp · known to Oktano sensor deployed — no AI usage observed3 Windows workstations sensorless by platform, not by choice — kamal-win-01 · rob-win-01 · kai-win-01known platform gap: sensor is macOS-only
kai-win still shows a detected-only Grok instance via egress — unsensored is not invisible, just unmonitored.
Peopleagents, devices, accounts, and flow per person — agent attributes are filters here, not rows
tokens ⇄ dollars — "both are just ways of gauging the flow"
agent-attribute filters land here from Inventory's drill — try ?tech=claude-code →
Team | Devices | Agents | Account | Tokens · 30d | Intensity | What they do (LLM summary) | |
|---|---|---|---|---|---|---|---|
TB Tyler Brooks | Engineering | ●tyler-mbp | Claude CodeCodex CLI Anthropic · OpenAI | enterprise | 58M | ▮▮▮ heavy | Uses 2 harnesses almost entirely for code generation in qpoint/qtap and qpoint/qplane — heaviest single user on the fleet. |
KR Kamal Reddy | Engineering | ●kamal-mbp ○kamal-win-01 win | Claude CodeCodex CLIGemini CLI Anthropic · OpenAI · Google | enterprise | 52M | ▮▮▮ heavy | Uses 3 harnesses, all for code generation in qpoint/qtap and qpoint/qtap-installer-win — the Windows sensor port. |
DW Dana Whitfield | GTM | ●dana-mbp | ChatGPT OpenAI | personal + ent dwhitfield.72@gmail.com | 42M | ▮▮▮ heavy | Heavy ChatGPT use for proposals and outbound copy — splits between the enterprise org and a personal account. |
MC Marcus Chen | Engineering | ●marcus-mbp | Claude CodeCodex CLIGrok Anthropic · OpenAI · xAI | enterprise + unknown acct | 34M | ▮▮▯ moderate | Leading the OpenAI experiment — moved from Claude Code to Codex CLI mid-July; now also trialing Grok. |
RC Rob Castellano | GTM | ●rob-mbp ○rob-win-01 win | Claude DesktopChatGPTGemini CLI Anthropic · OpenAI · Google | enterprise | 26M | ▮▮▯ moderate | Uses 3 harnesses primarily for document creation and sales workflow — proposals, call summaries, sequence drafts. |
PN Priya Natarajan | Engineering | ●priya-mbp | Claude CodeChatGPT Anthropic · OpenAI | enterprise | 24M | ▮▮▯ moderate | Splits between Claude Code for qplane frontend work and ChatGPT for design-doc drafting. |
KN Kai Nakamura | Ops | ●kai-mbp ○kai-win-01 win | Claude DesktopGrok Anthropic · xAI | enterprise + unknown acct | 22M | ▮▮▯ moderate | Claude Desktop for vendor and finance workflows — plus Grok detected on his unsensored Windows workstation. |
JF Jonah Fitz | Engineering | ●jonah-mbp | Claude CodeHermes Anthropic · Nous Research | enterprise | 9M | ▮▯▯ light | Light Claude Code use on qtap tooling — plus a locally-hosted Hermes model detected on his machine (no account, unmonitored). |
EW Erin Walsh | GTM | ●erin-mbp | ChatGPT OpenAI | enterprise | 8M | ▮▯▯ light | ChatGPT for competitive research digests and battlecard updates. |
LG Leah Goldman | Ops | ●leah-mbp | ChatGPTGemini CLI OpenAI · Google | enterprise | 7M | ▮▯▯ light | Gemini CLI scripts for IT asset reports; occasional ChatGPT for policy drafts. |
SA Sofia Alvarez | GTM | ●sofia-mbp | Claude Desktop Anthropic | enterprise | 6M | ▮▯▯ light | Light Claude Desktop use for event copy and campaign briefs. |
NP Natalie Pierce | Ops | ●natalie-mbp | Claude Desktop Anthropic | enterprise | 5M | ▮▯▯ light | Light Claude Desktop use for meeting summaries and facilities planning docs. |
AR Alexis Romeroinvestigate | Engineering | ○alexis-mbp | — | seat · no telemetry | 0 | ▯▯▯ none | No sensor telemetry — but holds an active seat on the Anthropic enterprise account. Coverage gap: investigate. |
MT Miguel Torres | GTM | ○miguel-mbp | — | — | 0 | ▯▯▯ none | No sensor on device — no AI usage observed. |
OH Omar Haddad | Ops | ○omar-mbp | — | — | 0 | ▯▯▯ none | No sensor on device — no AI usage observed. |
15 people — zero/low-usage rows kept visible: "who's not using AI very much is still interesting to management"
● sensor installed · ○ no sensor (amber) · dashed agent chip = detected-only · row click-through → user detail; a tech click there lands in Inventory double-filtered (the inverse of this page's ?tech= chip)
1 instance with no person — Codex CLI under the org key at the egress collector, no managed device. Attribution debt: identity mapping won't always be obvious, so the unknown state is shown, not assumed away. resolve in Inventory →
Composition notes
- Decisions applied (call, settled): machines live HERE, out of Inventory; agent attributes appear only as filters (the URL-backed ?tech= chip is Inventory's drill landing); sessions belong to the person (expand card + user detail), not Inventory; monitored vs detected-only is first-class on both device dots and agent chips.
- Coverage denominator = identity source of truth, not the sensor's own sightings — the header leads with Okta/Jamf integration state (15 people / 18 devices known) and coverage is a fraction of that. Related open question, annotated in the header: does "fleet" mean sensored devices or the whole org?
- Open question — name TBD: "Users / Devices" vs "Teams & Users" vs "Personnel". Working title kept, annotated in the mock's title bar; not silently decided. Likewise "sensor" over "monitor" for the endpoint software — the call leaned sensor, this page follows without settling it.
- The uncovered strip is the page's verb. Alexis (enterprise seat, zero telemetry) renders as investigate-red; Miguel and Omar as deploy actions; the three Windows workstations are separated out as a platform gap (macOS-only sensor), with kai-win's detected-only Grok showing that unsensored ≠ invisible.
- Zero-usage rows are content, not noise — the roster keeps intensity-none people visible ("who's not using AI very much is still interesting to management"); the footer says so.
- Tokens ⇄ dollars toggle on one Usage column — "both are just ways of gauging the flow"; the column re-sorts on whichever is shown.
- Attribution edges shown: the unattributed Codex CLI instance gets the amber attribution-debt band (teams-ideas idiom), linking toward Inventory where the account/key evidence lives.
- Every number derives from coverageStats() / monitoredStats() / allDevices() / allInstances() / people in the shared c21 fixtures — nothing hand-typed; Rob's and Kamal's summary lines carry the call's verbatim phrasing.