Salesforce Capped AI at $1,000 a Month. Put Your Token Budget in the Offer
Salesforce, JPMorgan, Coinbase, and Tesla now cap staff AI spend. How to find the heavy users who feel those caps and win them with a token budget.
If you hire engineers, the monthly AI budget an employer gives each person is now something candidates can compare. The catch is that the people who feel the caps most are the hardest to find, because very few of them list the tools on their profiles.
Salesforce made the comparison concrete. On Monday, Nov 16, it told employees it would cap their AI spending by role, at between $100 and $1,000 a month. Business Insider published the memo the next day, the same week Salesforce reportedly started its fifth layoff round of 2026. The engineers who hit these caps are the ones worth calling.
What did Salesforce's AI token budget memo actually say?
The memo came from COO Miguel Milano. It set monthly AI spending caps by role, from $100 for non-technical sellers up to $1,000 for technical staff inside the sales organization, and it added an approval chain for anyone who needs more. Salesforce publicly disputes that these are companywide individual budgets.
Here is how the reported Salesforce AI token budget works:
- $100 a month for non-technical sellers
- $500 a month for technical delivery roles
- $1,000 a month for technical roles inside the sales org
- Up to 20% over the cap needs a second-line manager's approval
- Up to 3x the cap needs a third-line manager and an ROI justification, and the increase lasts 90 days
Milano gave the reason directly: "Last quarter, some individuals spent thousands of dollars per month on tokens." He also wrote that "with AI confidence now high, we're shifting focus from adoption to impact."
Salesforce disputes the framing. A spokesperson told Business Insider the company does not have companywide individual token budgets and that AI usage is left to team leaders. So don't open a message by telling a Salesforce engineer they're capped at $1,000. The $1,000 tier is for technical roles in sales, and engineering teams may run under different rules. Ask the candidate what their number is.
The background explains why finance moved. In May, Marc Benioff said Salesforce expected to spend roughly $300 million on Anthropic tokens this year, mostly on coding agents. Once a cost that large gets described in terms of "impact," budgets start getting handed out one person at a time.
How does the Salesforce cap compare with JPMorgan, Coinbase, and Tesla?
Converted to a monthly figure, the reported caps run from $100 at Salesforce's non-technical sales tier to more than $21,000 at the top of Coinbase's weekly range. That spread is why the token budget now works as a comp line candidates can compare.
| Employer / tier | Stated cap | Monthly equivalent |
|---|---|---|
| Salesforce, non-technical sellers | $100/mo | $100 |
| Salesforce, technical delivery | $500/mo | $500 |
| Tesla | $200/wk | ~$867 |
| Salesforce, sales-org technical | $1,000/mo (+20% or 3x with approval) | $1,000 / $1,200 / $3,000 |
| JPMorgan (some Claude Code users) | $2,000/mo | $2,000 |
| Coinbase | $500 to $5,000/wk | ~$2,167 to ~$21,667 |
Caps from Business Insider and Bloomberg reporting. Monthly figures are converted from weekly caps (×52/12).
The rest of the market:
- JPMorgan started in September. When an engineer hits the limit, they get a new error that reads "ExceededBudget" and "Budget=2000.0."
- Apple cut some engineers' daily AI budget from $300 to $200, according to Bloomberg.
- Atlassian gives R&D staff capped "AI wallets" of $500 to $2,000 a month.
- Ramp, Valon, and n8n go the other way. Ramp advertises "Unlimited AI token usage," and n8n says "Everyone gets an unlimited AI budget."
Why do token caps hit your best engineers first?
A cap only limits the heaviest users, and the heaviest users tend to be the most AI-native engineers. Most employees never get near the limit. The people who do are the ones a startup should want most.
Salesforce's own pilot shows this. Spending in the second month came in significantly below budget, and output did not drop. For the typical employee, then, the cap changes nothing. For the person who was spending thousands a month, it adds steps: a 20% request to a second-line manager, then a 3x request with an ROI justification to a third-line manager, and that increase expires after 90 days. All of that friction lands on exactly the work that made them productive in the first place.
A token cap goes unnoticed by your median engineer. Your best engineer notices it the first week.
Jensen Huang offers the opposite benchmark. He said he would be "deeply alarmed" if a $500,000 engineer spent only $5,000 worth of tokens in a year. Salesforce's highest standard tier of $1,000 a month comes to $12,000 a year, a little over twice the figure Huang called alarming, and nowhere near the half-of-base-salary token budget his comment points toward.
Which roles feel the $1,000 cap hardest?
The technical roles inside Salesforce's sales org feel it most: solutions engineers and the people who build Agentforce demos for customers. Their daily work depends on AI, and the product they sell is AI agents.
A solutions engineer is a technical seller who builds demos, proofs of concept, and integrations so a deal closes. At Salesforce right now, that means building Agentforce demos for prospects, which takes a lot of prompting, re-running, and debugging agent behavior. When the company selling AI agents rations AI for the people who demo them, you have a very easy pitch: "You sell agents. I won't ration yours."
When recruiting Salesforce engineers from this group, search for these first:
- Solutions Engineer and Senior Solutions Engineer
- Adjacent technical sales titles such as Sales Engineer and Solution Architect
- Technical delivery and implementation roles (the $500 tier)
- Anyone whose profile or posts mention building Agentforce demos or proofs of concept
The closest role at a startup is the forward deployed engineer: an engineer who works inside a customer's environment to ship the product. It's the same work with more code and, if you write the offer well, a bigger token budget.
Why does searching for "Claude Code" find the wrong people?
In the US, more than a third of profiles that list Claude Code as a skill belong to founders and executives, not the engineers running up token bills. Those heavy users rarely put the tool on their profile, so a keyword search returns people with the right vocabulary but the wrong job.
In Refolk's index of professional profiles, 7,564 US profiles list Claude Code as a skill, compared with 3,674 in India, so the US has 2.06x as many. Who those people are differs a lot by country:
| US | India | |
|---|---|---|
| Profiles listing Claude Code | 7,564 | 3,674 |
| Who shows up (25-profile sample) | 9 of 25 founders, C-level, or VP | Mostly ICs: GTM Engineer, Software Engineer, Staff SWE, Tech Lead |
| Top region in sample | Austin (6 of 25) | Bengaluru (10 of 25) |
The reason is simple. Putting a tool on a profile is a way of marketing yourself. A founder lists Claude Code to show the company is AI-forward. An engineer who uses it eight hours a day treats it as part of the furniture and doesn't think to mention it. So the people who use the tool most are the least likely to say so.
Combining Salesforce-related keywords with engineering titles was worse: that query in the index returned 0 matches. A search by keyword turns up executives, a search by employer turns up nobody, and the remaining option is a sourcer reading repos one by one for weeks while the req stays open.
Where do heavy AI-tool users actually show up?
They show up in their code. GitHub repos with a CLAUDE.md file and commits with Claude co-author trailers tell you far more than any skill listed on a profile.
The signals worth searching for:
- CLAUDE.md files. A CLAUDE.md is a project instructions file for Claude Code. If someone has written one and kept updating it, they use the tool on that codebase all the time.
- Co-author trailers. Claude Code can add a Claude co-author line to commits it helps write. Check whether that is still the default before you build a search around it, because users can turn it off.
- Checked-in agent configuration. Repos where someone maintains agent setup alongside the code show a person building workflows on the tool, not just trying it out.
- Open-web writing about cost. Posts and threads about token bills, model routing, or caching come from people who have spent enough to care about the cost.
A GitHub signal by itself doesn't tell you who's capped. You also need the person's current employer, their role, and whether they're in Salesforce's sales org or at another capped company. Matching GitHub accounts to profiles and the open web is the slow part, and it's the part Refolk handles: you describe the person in plain English and get back a ranked shortlist.
How should you put a token budget in outreach and offers?
Give a specific monthly number, name the models it covers, and say who approves anything above it. A concrete figure beats "unlimited" because the candidate can put it next to Salesforce's 20% and 3x approval chain and compare line by line.
"Unlimited" sounds generous, but a candidate who just watched Salesforce, JPMorgan, and Apple bring in caps has reason to doubt it will survive the next budget review. There's a less obvious catch too. Coinbase cut its AI bill by nearly 50% without capping usage. It made open-weight models the default at $1.40 per million tokens, compared with $5 for Opus, and raised cache hit rates from 5% to 60%. That's good engineering, but it shows that "unlimited" can quietly mean "unlimited on a cheaper default model." A heavy user will ask which models they get.
Setting a Claude Code budget per engineer for an offer:
- Pick a monthly number. For example, "$3,000 a month." That matches Salesforce's top 3x tier, without the ROI justification.
- Name the models. Say "frontier models, including Opus." The engineer is comparing you to something specific.
- Define who approves overages. "No approval below $3,000; your manager signs above it" is one step. Salesforce's process has two levels plus a business case.
- Make it permanent. Salesforce's 3x increase lasts 90 days. Yours shouldn't expire.
- Write it into the offer letter as its own line next to the equipment stipend, so it reads as compensation rather than a perk that can be withdrawn.
An outreach message can be short:
Saw your CLAUDE.md work on [repo]. Solutions engineers here get $3,000/month in frontier-model tokens, no approval chain, written into the offer. Worth 20 minutes?
That message mentions a real artifact the person built, a specific dollar figure, and a contrast they'll recognize, and it does all three before mentioning the company.
Should you target laid-off Salesforce staff or the ones still there?
Target the ones still there. Eligible US employees laid off by Salesforce can get up to 30 weeks of base pay in severance, so they have little reason to rush. Employees who are still working under a cap run into it every week.
What's known about the Salesforce layoffs in November 2026 is limited. CXM reports this is the company's fifth round in 2026. Salesforce hasn't said how many people are affected, or which teams or locations. Goodall Research disagrees with that framing and describes the 2026 cuts as targeted and weighted toward senior roles, not a company-wide round. Either way, people who have kept their jobs through repeated cuts and now also face a new AI budget are a receptive audience.
How to work this pipeline:
- Contact capped employees now. They are doing heavy AI work this month, and they're starting to compare budgets.
- Keep warm ties with laid-off staff for later. With up to 30 weeks of severance, many won't commit until near the end of that period. Check in rather than pushing for a quick decision.
- Rank by GitHub signal, not seniority. If the cuts were weighted to senior roles, many of the most senior people may already be gone, and the code signals are what show you who the heavy users are.
FAQ
Is every Salesforce engineer capped at $1,000 a month?
No. The reported $1,000 tier covers technical roles inside the sales organization. Technical delivery roles are at $500 and non-technical sellers at $100. Salesforce also told Business Insider it has no companywide individual token budgets and leaves AI usage to team leaders. Ask each candidate what their actual limit is instead of assuming.
What is a competitive AI token budget for engineers right now?
Published reference points include Atlassian's $500 to $2,000 monthly AI wallets, JPMorgan's $2,000 a month for some Claude Code users, and Coinbase's $500 to $5,000 a week. A specific figure at or above the top of the range the candidate currently has, with no approval process below it, is easy for them to compare and believe. "Unlimited," as Ramp, Valon, and n8n advertise it, sounds bigger but is harder to compare.
How do I find engineers who use Claude Code heavily?
Look at code, not profiles. In Refolk's index, 36% of a US sample of profiles listing Claude Code were founders, C-level, or VP, and a Salesforce-plus-engineering query returned zero matches. Search GitHub for CLAUDE.md files and Claude co-author trailers on commits, then match those accounts to current employers and roles.
The data behind this
Try it on the search you came here for
Stop building boolean strings. Just describe the person.
Type one sentence. I plan the search, read GitHub, public LinkedIn and Crunchbase records, and the open web as it is right now, and hand back a ranked list with the reason next to every name.
01Describe them
One plain sentence. Role, city, stack, stage, whatever matters to you.
02I read the web live
GitHub, public LinkedIn and Crunchbase records, the open web. Not a database that went stale last quarter.
03You read the shortlist
Ranked, with the reasoning under every name. Open a profile, ask a follow-up, narrow it down.
- Staff backend engineers in NYC who shipped Rust in production
- Series A fintechs in SF under 50 people, growing headcount this year
- Maintainers of fast-growing Rust web frameworks on GitHub
- No boolean, no filters, no seat to buy. One box.
- Read at search time, so a profile updated yesterday counts today.
- Every step visible as it runs, every name with its reason.
500 free credits on sign-up. No card, no demo call. See real searches.