The Wired Connection faster Updated — Leads only

Automation team · internal

Seat Usage · Claude Enterprise

Who is using the Claude seats and how heavily: chat, Cowork, Claude Code, and the Office agents, per person and per product. Figures are usage credits, billed monthly on top of the seat fees: the companion bill to invoiced API-key spend.

—
Scope: these figures are Claude Enterprise seat usage only: the claude.ai apps, Cowork, and Claude Code signed in with an TWC seat. Dollar amounts are usage credits: metered usage beyond each seat's included allotment, billed to TWC monthly in arrears at standard API rates (per Anthropic's billing docs; invoice reconciliation pending). They are the numbers the spend caps and credit requests govern. The Console API keys are a second, separate metered bill, tracked on Token Governance. Reminder for the adoption push: Cowork is not a BAA-covered surface, so no PHI in Cowork.

Billing model · verified against Anthropic's billing docs and the signed contract

Two bills: the seat plan is prepaid, and seat usage credits are billed monthly on top.

Per Anthropic's documentation and Section 3 of our own Order Form, a seat buys access plus a bounded included allotment; usage beyond it draws usage credits i, which are billed monthly in arrears at standard API rates, the same per-token prices the API keys pay. The dollar figures on this page are that billed spend. Moving a person between an API key and a seat moves the bill; it does not shrink it. What seats do buy: HIPAA-Ready Claude Code with zero data retention, per-person monthly caps, and one approval queue.

Two real bills

Seat plan $17,580/yr prepaid, plus usage credits billed on top.

July so far: $10,470 of seat usage credits billed, projecting ≈$31.6K by month-end. Trailing 91 days: $75,731. The API keys billed $19,570 in the last 31 days on top of that. All of it is real money.

The rule that stands

People use seats. Robots use API keys.

Still the right policy, for compliance and control rather than savings: the Chat+Code seat is the HIPAA-Ready/ZDR SKU, and seat spend sits under per-person caps and one approval queue. A raw key without ZDR is not a covered surface.

Where savings actually are

Model right-sizing, idle seats, dead keys, caps.

About a quarter of seat-side spend rides legacy Opus models; leaver and idle seats are pure waste; unused keys are risk. Those levers cut real spend on both pipes. Switching pipes does not.

API keys, last 31 days
$19,570 invoiced
Seat usage credits, monthly pace
≈$25.3K–$31.6K billed

Seat pace: $75,731 over the trailing 91 days (≈$25.3K a month) with July projecting ≈$31.6K. Per Anthropic's docs the spend export reflects usage beyond seat allotments, billed monthly in arrears; final confirmation is the org's Billing page and invoices, which is move 1 below.

This week's four moves

  1. Get the invoices. An org Owner pulls the claude.ai Billing page and the monthly usage-credit invoices across all three Anthropic billing accounts; the Automation team reconciles them against this page's totals. This is the artifact that closes the last open question on this page. confirms the corrected mathorg Owner + Automation team
  2. Reclaim the dead seats. Six seats still assigned to people who left, plus five more idle ones; ten minutes in the claude.ai admin console, and seat removal joins the offboarding checklist. cuts the 26-seat overage in halfany org Owner
  3. Right-size the legacy-model users. Roughly $20K of the last 90 days of seat usage sits on two superseded Opus versions. Now that seat usage is a monthly bill, nudging those users to current Sonnet-class models is the largest honest savings lever on this page. real monthly moneyAutomation team
  4. Rein in the card account. A second Anthropic account ("TWC Care") is auto-charging ~$90 a pop to a card, twelve times since June 20 and accelerating; put it under governance or shut it off. stops an ungoverned burnAutomation team + finance
See the mechanics and the word-for-word receipts ↓
Reading the dollar figures on this page? One thing first. They are real spend: usage credits billed monthly in arrears at standard API rates, on top of the fixed $17,580/yr seat contract (Anthropic's billing docs plus Order Form Section 3; the org's invoices are being pulled to reconcile). The full billing mechanics live in the banner above and the two-pipes panel.

The corrected case: seats are for control and compliance, not free usage i

people on seats, keys only for machines · corrected 2026-07-15
People on pay-per-use keys i
$18,244
of the $19,570 API bill, last 31 days (93%)
Seat credits billed, July so far i
$10,470
billed monthly in arrears · projecting ≈$31.6K by month-end
The fixed seat contract i
$17,580 /yr
75 seats, 34 with Claude Code · fee prepaid to 2027-03-16; usage credits bill on top
Per-token rate gap between pipes i
≈$0
credits bill at standard API rates; US-only inference may add 1.1x
The seat numbers on this page, decoded. Three different counts appear on this page and they are easy to blur together. 75 is what TWC bought: 75 seats i for $17,580 a year, prepaid to 2027-03-16. Those 75 come in two types: 41 are chat-only seats ($15 a month) and 34 are seats that also include Claude Code ($25 a month). 101 is how many people currently hold a seat, which is 26 more than we bought; how Anthropic bills those extra assignments is an open question for our rep (step 1 below), not something this page can answer. So when the compliance note says "34 Claude Code seats," that is the Claude Code portion of the 75, not a different contract.

And to be exact about what TWC actually pays Anthropic, there are four money streams, not one. (1) The seat plan: $17,580 a year, fee prepaid, verified against the signed Order Form. (2) The seat usage credits: metered usage beyond seat allotments, billed monthly in arrears at standard API rates; $75,731 over the trailing 91 days per the org's own spend export, which makes this the largest stream at the current pace (invoice reconciliation pending). (3) The Console API keys: $56,401 invoiced between 2026-02-03 and 2026-07-09, verified against Anthropic's cost reports. (4) Smaller streams still being verified line by line against invoices: the 26 seats assigned beyond the 75 (the contract bills extra seats monthly at the listed per-seat price), and a series of ~$90 "auto-recharge credits" card charges hitting a second Anthropic billing account named "TWC Care" (12 receipts since June 20, about $1,085, accelerating in July; these invoices go to an individual inbox and sit outside both this page's data feeds and the normal finance loop). An earlier version of this note called the seat-side usage charges a "smaller stream"; the export says otherwise.
What is happening. TWC pays for Claude through two metered pipes plus a fixed fee. The seat plan is $17,580 a year for 75 licenses; on top of it, seat usage beyond each seat's included allotment bills monthly in arrears as usage credits at standard API rates ($75,731 over the trailing 91 days). Separately, API keys i bill per token the same way: $19,570 in the last 31 days, $18,244 of it people in Claude Code sessions rather than systems. An earlier version of this callout said the seats "would mostly have covered" the key spend; that was wrong. The same work bills at about the same per-token rates on either pipe.
How the two pipes should be used. A person signs into Claude Code with their TWC seat, the same login as claude.ai. Their work draws first on the seat's included allotment, then bills as usage credits through extra usage i at the same per-token rates the API charges, under a monthly cap that we set. API keys remain for machines: the n8n automations, the contract tools, the scheduled jobs. People get seats; systems get keys, because seats carry the HIPAA-Ready/ZDR terms, the caps, and the approval queue, not because seat usage is cheaper.
The receipts: what the signed paperwork actually says. These are word-for-word quotes from the three governing documents, all in the company Drive. Nothing below is interpretation.
Enterprise - Chat (HIPAA-Ready) and Claude Code  ·  $25.00, per User seat, per month  ·  34 Anthropic Enterprise Order Form, signed 2026-03-17 · the seats TWC bought
Claude Code with Zero Data Retention: Covered under BAA; ZDR enabled automatically when purchasing the Claude Code with ZDR SKU … no additional configuration required. Anthropic "HIPAA-Ready Claude Enterprise" coverage guide, March 2026 · that SKU is the one above
$25 / seat / month + pooled usage  ·  Enterprise - Chat (HIPAA-Ready) and Claude Code (ZDR) Same coverage guide, pricing box · "pooled usage" is the built-in overflow
Charges are calculated based on actual token usage at Anthropic's then-current API pricing rates … Customer may configure spending limits at the organization level and/or per User through the administrative console … invoiced monthly in arrears. Order Form, Section 3 (Usage) · how overflow is billed and capped
Put together: the seats TWC pays for include HIPAA-covered Claude Code and a built-in, capped overflow plan billed at the same rates the API charges, invoiced monthly in arrears. That last clause is the correction: overflow is a real bill. The paperwork makes the compliance-and-control case for seats; it does not make a savings case.
Compliance check, verified against the signed agreements. The 34 Claude Code seats (the $25 type in the decoder above) are the HIPAA-Ready SKU with zero data retention i built in, a covered surface under the Anthropic BAA i (signed March 2026). A plain API key without zero data retention is not a covered surface. So moving a person from a key to a Claude Code seat is equal or better for PHI compliance, not a tradeoff. Two footnotes: the restricted Fable research workspace stays on keys and stays PHI-free by policy, and we will ask Anthropic to confirm in writing that overflow billing does not change a seat's data handling.
The go-forward path.
  • 1 · Get it in writing from our Anthropic rep (MG Carroll). The overflow system is already on and working: the per-person caps, the pause at the cap, and the approval queue shown on this page are that system operating today. Nothing needs activating. What we need on paper: (a) overflow keeps each seat's HIPAA terms; (b) how the seat count settles (101 assigned against 75 purchased, and how the extra 26 bill); (c) the roster of who holds which seat type, because people migrating off keys need the Claude Code type and the roster decides how many of the 34 are free; (d) what each seat's included allotment actually is, in dollars, and whether unused allotments pool across people; (e) copies of every Anthropic invoice across all three billing accounts (the Enterprise seat org, the Console API org, and the second "TWC Care" credits account) so total Anthropic spend becomes one verified number, and so this page's corrected billing model is confirmed against real invoices.
  • 2 · Ship the new setup path. The Automation team rebuilds the installer so a new machine signs into the person's seat and no API key lands in anyone's terminal. Existing key holders get a five-minute migration.
  • 3 · Close the gaps. Two active Claude Code users have no seat today; reclaim the six seats still assigned to people who left (see the disconnect audit below), reassign two to them (as Claude Code type, part of the step-1 roster check), and add seat removal to the offboarding checklist. Start the migration with the heaviest key users so their spend lands under seat caps and ZDR terms; expect the bill to move with them, not shrink.
The numbers behind this recommendation
  • Window and source. Invoiced API spend from Anthropic's own cost reports, 2026-03-10 through 2026-07-09: $56,401 total. The workspace totals reconcile to the cent against the comparison feed on this page.
  • How people were separated from systems. By usage fingerprint. Interactive Claude Code sessions read 80 to 92% of their tokens from cache i; machine workloads read near zero. By that test roughly $51,100 of the $56,401 was human Claude Code work; the conservative cut is $38,100.
  • The one flag. The largest single key (Leadership workspace, registered to kpemberton@) shows a machine pattern, near zero cache reads, so it may be a connected system that genuinely needs a key. It is excluded from the conservative figure and should be identified before anything moves.
  • Would the seats have absorbed it? Withdrawn 2026-07-15. This bullet previously claimed single seats had "absorbed $2,300 per month of Claude Code with no extra charges." Per Anthropic's docs the spend export only shows usage beyond a seat's allotment, so figures like that were billed, not absorbed. Allotment sizes are unpublished; whatever a seat does not absorb bills at the same per-token rates through extra usage, so moving work between pipes is roughly cost-neutral, and no savings figure can honestly be attached to it.
  • Freshness. Anthropic can restate the most recent ~30 days of cost data upward, so July figures may grow slightly.

Both pipes, real dollars i

corrected 2026-07-15 · the earlier "priced both ways" savings table is withdrawn
API keys, last 31 days i
$19,570
invoiced per key, from Anthropic's cost report
Seat usage credits, July so far i
$10,470
billed monthly in arrears · projecting ≈$31.6K by month-end
Seat fees i
$1,465/mo
$17,580/yr for 75 seats, prepaid to 2027-03-16
Combined pace i
≈$46K–$53K/mo
both metered pipes plus seat fees, at current pace
Correction, 2026-07-15. This panel previously priced the same month "both ways" and claimed $11,943 to $14,229 of monthly savings from signing API-key users into their seats. That math assumed seat usage was free after the seat fee. Anthropic's billing documentation and Section 3 of the signed Order Form say otherwise: seat usage beyond each seat's included allotment bills monthly in arrears at standard API rates, and the spend export behind this page reflects that billed overage. Because the same tokens bill at about the same rates on either pipe, no honest savings number can be attached to the switch; the seat allotments that would create one are unpublished. The claim is withdrawn, not restated. Confirming artifact: the org's Billing page and monthly invoices (move 1 in the banner above).
The disconnect audit: 13 zero-usage seats, checked against the HR system (2026-07-10). i Six belong to people who have left the company: Giustino Veri (VP, Customer Success), Ryan Sousa (Director, Customer Success), Sharry Bain-McPhee (Chief of Staff), Kyle Bell (Sr. CSM), Arshad "TJ" Ahmady (Sr. AE), and Miles Fenwick (VP, Information Technology). Four are active employees who simply have not used the seat (Dan Uniejewski, Ilona Kozak, Ivan Kukharenko, Kwame McDonald), and one is an external address we could not match (jcastellan@braincloudgroup.com). The remaining two, Jay Hunt and David Thackeray, are not actually idle people: they work in Claude every day on metered API keys, which is exactly the habit this page recommends ending. Migrate those two; do not reclaim their seats. Reclaiming the other eleven returns no cash mid-term (the seat fees are prepaid), but it cuts the 26-seat overage exposure nearly in half, frees seats for the migration, and closes an offboarding gap: seat removal should be on the departure checklist.
Should we buy cheap seats to pad the pool? Not yet, and probably not. i The overflow pool bills at the same per-token rates as the API keys, so there is no rate arbitrage to capture: a padding seat only pays for itself if unused seat allowances pool across people AND each seat's included allowance is worth more than the seat costs. Neither is confirmed, and under seat-based plans allowances are usually per-person. The right order, in telecom terms: disconnect the six dead lines, move the metered people onto the plan, measure one clean month of real overflow, and only then size any seat purchase against it. At $25 a month, a Claude Code seat pays for itself the moment a real person uses it, so if sustained overflow shows up, the answer is more seats for more people, not padding. One question is now on the Anthropic list: do unused allowances pool org-wide, and what is each seat's included allowance in dollars? If the answer is "yes, and it is generous," revisit padding; get it in writing either way.

Active users i

Seat spend, day by day i

usage credits, billed monthly in arrears at standard API rates

Where the spend goes i

Spend by person i

Seat figures are usage credits, billed monthly in arrears at standard API rates (per Anthropic's docs; invoice reconciliation pending). Heavy Claude Code users look huge because roughly 90% of their tokens are cache reads, billed at about a tenth of the input rate but in enormous volume. The Token Governance page tracks the separate API-key bill.
Person i Products i Chat i Code i Cowork i Other i Total i ▾ Month vs cap i
How these numbers are made · plain-English definitions
  • Where the data comes from. Anthropic publishes two reporting feeds and we pull both into our own archive: the Enterprise Analytics feed for everything seat holders do in claude.ai (chat, Cowork, Claude Code), and the Console cost report for API keys. This page cannot see conversation content, only counts, tokens, and dollars.
  • Active people. Every activity count on this page counts distinct people, not clicks or sessions. "Active yesterday" means the person did anything at all yesterday. "Weekly (7d)" means they did something at least once in the trailing 7 days; "Monthly (30d)" is the same over 30 days. One message counts, so these measure reach, not depth.
  • Usage credits are a bill (corrected 2026-07-15). Seat usage is metered at standard API rates and, past each seat's included allotment, billed to TWC monthly in arrears (Anthropic help center, "Manage usage credits for Team and seat-based Enterprise plans"; Order Form Section 3). Anthropic's docs also state the spend export behind this page "only reflects spend that exceeds your seat allotment," so the dollars here are billed spend, not a notional valuation. An earlier version of this page called them "not a bill"; that was wrong. Remaining check: the org's Billing page and monthly invoices, being pulled to reconcile.
  • Why some people show thousands of dollars. Tools like Claude Code and Cowork re-read the whole working context on every step ("cache reads"). Cache reads bill at roughly a tenth of the normal input rate, but heavy sessions read billions of tokens, so they still add up to thousands of billed dollars on a roughly $20/month seat. Heavy use can be worth every cent; it is just not free, so model choice and habits matter on seats exactly as they do on API keys.
  • Reclaim and Gemini calls. "Reclaim" means the seat did nothing for 30+ days. "Move to Gemini" means the person only chats and uses none of the Claude-specific features (projects, skills, connectors), so Gemini, already included in our Google plan, covers them at no extra cost.
  • Trust checks. The per-person totals were reconciled against Anthropic's own admin export on 2026-07-10 and matched to the cent. Contract facts come from the signed Order Form. Data starts 2026-02-03 (first recorded usage) and runs through yesterday.
  • The impact ledger. Each run of the data pipeline saves a snapshot of the state of the world: who holds a seat, which seats are idle, each person's main model and 30-day mix, every spending cap, and every Console API key's status and trailing-31-day spend. The next run compares snapshots and logs any change this page had flagged, valued as a monthly run rate with the formula shown on hover. Only changes detected after Jul 13, 2026 earn credit, and credit means the page surfaced it while a person made the change; the page never changes anything itself. The page's own serving cost is computed from measured file sizes and traffic (estimated where per-page logs do not exist, and labeled as such). Values are run rates, not realized savings: a change must hold for the money to be real.