On November 2, 2026, Jason Calacanis posted a line that raced across founder timelines: he was spending under $40K a year on AI tokens and subscriptions and getting more than $2M in value back at Launch and This Week in Startups. He said the stack had automated at least seven full-time jobs and made average tasks roughly twice as fast. The post named Anthropic, Lovable, Grok Bot, ElevenLabs and a long tail of other products.
The quote is real. The temptation for a 15-person HVAC, dental, legal, or agency shop is to treat it like a shopping target. That is the wrong move. Kay's follow-up pushed the useful reading: the headline multiple is less important than the claim that specific jobs got automated and cycle time dropped. Your job is to steal the measurement habit, not the dollar figure.
TL;DR
- On November 2, 2026, Jason Calacanis wrote that under $40K a year in AI tokens and subscriptions was returning $2M+ in value and had automated at least seven full-time jobs at his companies — his numbers, not yours.
- For most owner-led service firms, the useful question is not how to copy that stack. It is how much to spend on one workflow before the spend becomes a hobby.
- Start with a monthly envelope tied to a timed task, include human review, and treat seat sprawl as a cost, not a status symbol.
- AI drafts, your team approves. The approval minutes belong in the budget.
What the $40K claim actually says
Calacanis described a media and investing operation that already lives in documents, audio, research and software. Those businesses burn tokens differently than a field-service company whose money moves through Jobber, ServiceTitan, QuickBooks and a phone that rings during storms.
He also framed the last 90 days as unusually productive. That is a personal operating update, not a benchmark study. When you cite it inside your own planning meeting, label it that way. Otherwise someone will treat a viral post as a finance policy.
How much should a small business spend on AI tools?
A practical answer for a 5–75 person service firm is: enough to clear one painful weekly workflow after review costs, and not a dollar more until that proof exists.
Public guides aimed at small businesses in 2026 often land in the low hundreds of dollars per month for a few targeted tools once you leave free tiers, with higher totals as seats multiply. Xero's small-business AI guide describes a phased start in the $50–$150 per month range for many firms. That range is a planning hint, not a rule. Your ceiling should come from the value of hours you can actually free.
Use this sequence:
- Pick one recurring job with a clear output: quote packs, intake summaries, invoice coding drafts, after-hours call notes.
- Time five runs the old way, including the messy exceptions.
- Time five runs with AI drafts plus human approval.
- Price the hours at a loaded internal rate you already use for pricing jobs.
- Subtract subscription and usage costs.
- Only then set a monthly envelope for that workflow.
If the math is thin, do not "invest more to make it work." Narrow the task or stop.
A starter budget table for a 12-person service firm
| Line | Monthly |
|---|---|
| One team assistant seat for the workflow owner | $20–$30 |
| One automation or voice add-on tied to the same workflow | $0–$80 |
| Estimated usage overage buffer | $20 |
| Owner review time (4 hours at $75 loaded) | $300 |
| Staff review time (6 hours at $45 loaded) | $270 |
| Working total to defend | about $610–$700 |
That table will look expensive next to a $20 chat subscription screenshot. Good. The screenshot usually hides the review labor. If AI drafts, your team approves, the approval is part of the product cost.
Compare that to copying a $40K annual stack. $40K is about $3,300 a month before your people touch anything. For many local service firms that number only makes sense after several workflows have already cleared a timed test.
Where small teams overspend
They buy a seat for everyone "so nobody feels left out." Occasional users stay on free tiers for non-sensitive work.
They pay for three writing tools that all draft the same email.
They keep an annual plan after the pilot person leaves and nobody owns the workflow.
They ignore usage meters on per-resolution support bots until the bill spikes in a busy month.
They celebrate draft speed and forget that a wrong invoice code still creates cleanup work.
How to read celebrity AI spend without getting rolled
Ask four questions every time a big account posts a stack cost:
- What industry and margin structure is this person in?
- Which jobs were timed before and after?
- Who still approves money, legal language, or customer promises?
- What would break if the tool vanished tomorrow?
If the post cannot answer those, it is entertainment. Enjoy it. Do not paste it into your bookkeeping software as a target.
A two-week spend pilot you can run without a consultant
Week one: freeze new AI purchases. List every AI-ish subscription and who uses it weekly. Kill obvious duplicates.
Week two: pick the single workflow with the loudest complaint. Run the timing test above. Write the result on one page: minutes saved, error rate, review minutes, monthly tool cost, go or no-go.
Share that page in your Monday meeting. If the room cannot explain the page, you are not ready for more spend.
Verdict: Hold on big AI budgets until one workflow pays
Verdict: Hold on matching public six-figure or high-five-figure AI budgets. Verdict: Buy a tightly scoped tool only after a timed pilot includes review labor. Verdict: Skip "seat for everyone" rollouts that exist to look modern.
The Calacanis post is useful because it treats AI like production software that should remove work. It is dangerous if you import the dollar amount without importing the measurement. Your shop needs a budget you can defend on a whiteboard, not a screenshot from someone else's company.
A ninety-day spending calendar
Month one: freeze new AI purchases except the timed pilot tool. Publish a one-page inventory of every AI-ish subscription, owner, monthly cost and last-used date. Cancel anything unused for thirty days.
Month two: run the timed pilot on one workflow. Include review labor. Write a go or no-go memo. If go, set a monthly envelope with a named owner.
Month three: expand seats only for people who touch the winning workflow weekly. Re-run the timing test. If savings vanished, shrink the envelope instead of adding tools.
This calendar feels slow next to social media screenshots. It is how shops avoid a December surprise bill and a drawer full of logins nobody remembers.
How to talk to your bookkeeper about AI
Bring invoices with product names highlighted. Map each to a workflow. Ask for a class in the chart of accounts so AI spend stops hiding inside software misc. Review that class in the monthly close. When finance can see the line, operators stop treating tokens like snacks.
Red flags in vendor pricing pages
Per-seat pricing that assumes everyone needs the highest tier. Usage meters with no hard cap. Annual prepay discounts that exceed your pilot horizon. Professional services bundled as mandatory. If you cannot model a bad month, you cannot approve the buy.
FAQ
Want a sane AI tool budget for your shop?
In a free 30-minute assessment we pick one workflow, price the current hours, and sketch a monthly AI envelope you can defend to yourself.
How much should a small business spend on AI tools?
Most owner-led service firms do well starting between a few hundred and a couple thousand dollars a year for a first workflow, not a $40K media-company stack. Tie the number to timed hours you can actually free, then raise it only after review quality holds.
What did Jason Calacanis say about AI spend?
In a November 2, 2026 post he said he was spending under $40K a year on tokens and subscriptions across Anthropic, Lovable, Grok Bot, ElevenLabs and other tools, getting $2M+ in value at Launch and This Week in Startups, and had automated at least seven full-time jobs while getting faster at average tasks.
Should my business copy a $40K AI budget?
No. That figure comes from media and investing companies with different margins and tool needs. Use the post as a reminder to measure your own return, not as a shopping list.
What belongs in an AI tool budget?
Subscriptions, usage overages, setup time, training, and the minutes people spend reviewing drafts before anything sends, posts, or pays.
How do I know if we are overspending?
If you cannot name the workflow each paid seat supports, or if review time erased the time savings, you are overspending. Cancel or pause before adding another tool.
Is a free AI plan enough?
Often for experiments. Move to paid seats when the free tier blocks a weekly job you already timed, not because a launch video said to.
How often should we revisit the budget?
Monthly for the first quarter of a pilot, then quarterly. Kill tools that do not clear a simple before-and-after test.