Issue #2: Moving from Claude Team to Enterprise? Every token gets billed at API rates

Hey,

This week Ask added four new questions about rolling out AI at work. Nobody has answered them yet, so the best reply to any of them could lead next week's issue.

From Ask

These are Started by Token Thrift topics. Each one opens with a summary of public sources (forum threads, vendor docs and standards) to get the discussion going. Replies come from readers under anonymous handles.

What changes when you move from Claude Team to Enterprise?

Anthropic's Team plan tops out at 150 seats. Enterprise adds audit logs, SCIM, a Compliance API and customer managed encryption keys, but the bigger change is billing: the seat covers access only, and all usage is billed at API rates. Anthropic's migration guide recommends upgrading your existing organization rather than starting a new one, and the self serve upgrade can't be reversed. Still open: how do you forecast monthly usage before the switch, and which spend limits stop surprises without locking people out?

Read the summary and share how your move went

How do agencies bill AI usage to clients?

A public thread describes a small agency that tagged coding sessions by repository and still could not match the totals to its invoice, because chat work and sessions with no git remote went untagged. Replies split between building AI into the rate and tracking every call per client with separate keys or a tagging proxy. Still open: do clients accept AI as a line item, or prefer it hidden in the rate?

Read the summary and share what you charge

How do you control built-in AI features in tools you already pay for?

Most AI at work now shows up as a button inside a tool you already license. Blocking AI sites one at a time was called whack a mole. Adobe's admin docs show AI in Acrobat can be switched off per product profile, but a user in two profiles gets it if either one allows it, and the browser extension has a separate setting. Still open: does anyone have a full list of which licensed tools have AI switched on by default?

Read the summary and share your approach

Also new: How should you evaluate AI agent observability tools? One concrete test from a public thread: replay 10 to 20 past incidents through each candidate and score whether it found the root cause.

Replies use a random handle, and names and employers are never shown. Got a question about rolling out AI at work? Ask it anonymously.

Deals this week

Every deal below was checked on the vendor's own page on Oct 28, 2026.

Ending soon

ElevenLabs Starter: $1 for the first month, until Nov 1

Normally $6 a month, with 30,000 credits a month. See ElevenLabs pricing

View this deal

Fireworks Muse Glimmer: free fine-tuning until Nov 30

Muse Glimmer 30B supervised fine-tuning is free in the UI and the Serverless Training API. No credit card required. See Fireworks pricing

View this deal

Deepgram Flux TTS match: up to $500 through Dec 31, 2026

For every $1 you spend on Flux TTS, Deepgram adds $1 in credits. Pay As You Go and Growth plans. See Deepgram pricing

View this deal

InVideo Starter: 50% off annual for first-time AI creators

$10 per seat a month billed annually, down from $20, with 400 credits per seat a month. Get InVideo Starter (affiliate link)

View this deal

New guide: Microsoft for Startups credits. Microsoft now publishes its starter amounts: if you're new to Azure, $2,000 for 90 days, then $4,800 more for 180 days after business verification, and up to $150K over time as you show progress and Azure usage. Read the guide

Everything else is in the deals directory, and all 15 startup credit programs are on the credits hub.

The InVideo link above is an affiliate link. If you sign up or buy through it, Token Thrift may earn a commission at no extra cost to you. The other deal links in this issue are not affiliate links.

Token Thrift

Deals checked against vendor pages. Always confirm terms on the vendor site before you sign up.

Privacy