Jack Dorsey just re-anchored the hiring bar for every AI-lean company that reads Block's shareholder letters, which is most of them. Gross profit per employee, not headcount growth or scope, is now the number that pulls a resume off the pile. If your bullets still lead with team size, tickets closed, or "shipped model X," you are going to lose a slot to someone who quantified the exact same work as revenue per head.
What Dorsey actually said, and why hiring managers copied it
Dorsey publicly targeted more than $2M in gross profit per employee at Block, roughly 4x the company's pre-COVID efficiency, after cutting headcount from 10,205 to just over 6,000. That single X post ("$2M+ gross profit per person, 4x our pre-covid efficiency, which stayed flat at ~$500k from 2019 until 2024") is the sentence hiring managers at AI-lean startups have been pasting into rubrics.
The mechanics are worth understanding before you rewrite a single bullet:
- Block's gross profit per employee rose from about $500K (2019 baseline) to roughly $750K in 2024 to about $1M in 2025, per CFO Amrita Ahuja.
- Ahuja told Fortune the metric could hit approximately $2M in 2026 if projections hold.
- Block deployed its open-source AI agent Goose to all 12,000 employees in eight weeks, with engineers reporting time savings of 8 to 10 hours per week and shipping 40% more code per person in six months.
- Dorsey warned publicly that most companies will face the same workforce reckoning within a year.
The reason the ratio spread so fast: it collapses "AI leverage" into one auditable number. Instead of asking "do you use Copilot?" a hiring manager asks "what is your GP-per-head contribution?" That is a question your resume has to answer in a bullet, not a phone screen.
$500K to $2M+, with headcount cut from 10,205 to under 6,000.
The new benchmark table every candidate should know
Revenue per employee at AI-native companies now runs 6 to 13x the public SaaS average, and that gap is why "senior engineer" bullets from 2022 read as bloated in 2026. If you are applying to a company that quotes any of the numbers below, your resume needs to speak in the same unit.
| Company / segment | Rev or GP per employee | Source |
|---|---|---|
| Block, 2019 baseline | ~$500K GP | Dorsey, Feb 27 2026 |
| Block, 2025 | ~$1.0M GP | Ahuja, CFO.com |
| Block, 2026 target | $2.0M+ GP | Dorsey shareholder letter |
| Klarna, 2024 to 2025 | $575K to ~$1.0M rev | TechCrunch, May 2025 |
| Gamma (50 people, $100M ARR) | ~$2.0M rev | TechCrunch, 2025 |
| AI-native startup median | $2.0M to $4.0M rev | Redpoint via Forbes |
| Public SaaS average | ~$300K rev | Redpoint via Forbes |
| Apple | $2.38M rev | Stocklytics 2025 |
Two derivations worth internalizing: Block's target is 6.7x the public-SaaS average, and its 4x per-head jump implies each retained engineer must absorb the output of roughly three laid-off peers (10,000 × $500K ≈ $5B; 6,000 × $2M = $12B). That is the math your bullet has to match.
Why the "4x" is not really 4x from AI, and why that matters for your baseline
Dorsey's 4x is measured from a post-bloat trough, not from Block's true pre-pandemic peak. In 2019 each Block employee generated closer to $678K in gross profit; the ~$500K baseline he cites reflects the 2022 hiring binge. The practical implication for your resume: quantify against your own pre-AI baseline, because that is the framing hiring managers now copy.
Concretely, your bullets should read as before/after inside your own team, not against an industry average nobody can verify:
- Weak: "Used Copilot and Cursor to accelerate feature delivery."
- Weak: "Shipped 40% more code (industry benchmark)."
- Strong: "Held team headcount at 6 while doubling monthly shipped features from 11 to 22, deferring two backfills at ~$220K fully loaded."
- Strong: "Rebuilt fraud-scoring pipeline in 3 weeks that a prior team had taken 2 quarters to build, freeing one SRE and one ML engineer for the payments launch."
The second pair works because it names a cost line removed, not a feature shipped. Ahuja specifically cited a Block risk underwriting model that previously took an entire quarter and was completed in a fraction of the time. That is the sentence template hiring managers are scanning for. Translating a messy work history into that exact voice, for every posting, is the tedious part; it is also the exact work Refolk takes off you. Paste the job description, get your own resume back rewritten in the units the company measures.
Five bullet patterns that clear a $2M-per-head rubric
The bullets that survive share a shape: unit of output, unit of headcount, unit of dollars, and a comparison to a prior baseline. Here are five templates that map onto the Dorsey framing.
- Compressed workflow. "Compressed [X-quarter workflow] to [Y weeks] using [named agent/tool], deferring [Z hires] at ~$[N]K loaded cost."
- Held headcount, doubled throughput. "Kept team at [N] engineers while [metric] grew from [A] to [B] over [period]."
- Cost line removed. "Retired [vendor/contract] worth $[N]/yr by shipping [internal replacement] in [weeks], reallocating [N] FTE to [revenue project]."
- Revenue per head attribution. "Owned [product surface] contributing $[N]M ARR across a team of [N], implying $[X]M revenue per engineer."
- AI-native rebuild. "Rebuilt [legacy system originally built by N people over M months] as sole owner directing [Claude Code / Goose / Cursor agents] in [weeks], at ~$[N] in inference cost."
The fifth pattern has a real-world anchor most interviewers know: Y Combinator's Garry Tan rebuilt his 2008 startup Posterous (originally 6 to 7 people, ~18 months, ~$4M raised) in January with about $200 of Claude Code, in five days, mostly one person directing agents. Tan calls the approach "tokenmaxxing." If you have done anything shaped like this, name the model, the cost, and the prior team size. Those three numbers are what makes the bullet believable.
Bullets that name a cost line removed beat bullets that name a feature shipped, every single time.
Where the AI-fluent senior engineer pool actually sits
In Refolk's index of US professionals tagged "Software Engineer" or "Staff Software Engineer" with AI/LLM skills, the pool is 321 profiles, concentrated at Meta (5), Google (5), Apple (4), plus Ambient.ai and Pinterest, with the San Francisco Bay Area accounting for 6 of the top 25. That is a small, tightly clustered pool, and it changes how you should write.
Two consequences follow directly:
- FAANG-vs-FAANG is the default matchup. A raw "shipped model X at Meta" line no longer differentiates when the next resume in the stack also says "shipped model X at Meta." The differentiator is the GP-per-head or throughput-per-head number attached to it.
- Geography leaks into the rubric. Bay Area hiring managers assume you have used Cursor, Claude Code, or an internal agent daily. If your resume treats "prompt engineering" as a skill line rather than an assumed baseline, you read as two years behind.
Rewriting 321 versions of the same resume for 321 postings by hand is why most candidates give up and send the generic one. Refolk drafts each tailored version from your actual history, scores how well you fit the posting, and writes the cover letter in the same voice, so the leverage bullet you spent an hour crafting lands in every application.
The Klarna counter-hook: how to not overclaim
Do not write bullets that say you replaced N humans with AI. Klarna publicly claimed its AI tools could do the work of 700 customer service agents and cut its employee base from 7,400 to 3,000, then quietly backtracked a year later. Every interviewer has read that story.
Bullets to avoid, and their safer rewrites:
| Overclaim (reads as red flag) | Safer version (reads as credible) |
|---|---|
| "Replaced 12 support agents with LLM triage." | "Deflected 38% of tier-1 tickets via LLM triage, letting the team hold at 12 headcount through a 2x ticket-volume quarter." |
| "Automated the analytics team's job." | "Cut recurring dashboard requests from 40/week to 6/week, freeing 1.5 analyst FTE for revenue modeling." |
| "10x engineer with Cursor." | "Shipped 40% more PRs per week vs. my own H1 baseline after adopting Cursor + Claude Code." |
The pattern: name what stayed constant (headcount, quality, SLAs) alongside what improved. That is the shape of a claim hiring managers can defend to their CFO.
The comp premium, and why this bullet is worth an afternoon
Rewriting these bullets is not cosmetic; it is a raise. Hunt Club's 2026 VC compensation benchmarks show leaders who have operationalized AI are capturing 10 to 15% higher base pay. On a $220K base, that is $22K to $33K a year, every year, from a paragraph of resume copy.
About 50 people, $100M ARR, profitable for years. The clean "1 person = $2M" example every AI-lean startup cites.
Vertical matters too. Median revenue per employee for vertical SaaS runs $187.5K vs. $158.7K for horizontal SaaS, a gap of nearly $30K per head. If you are targeting a vertical AI startup (fintech, legal, healthcare), benchmark your bullets against the higher bar; if you are targeting a horizontal tool company, the ~$300K public-SaaS number is closer to what they measure against internally.
The 30-minute rewrite, in order
Do this in one sitting, top to bottom, before you send another application:
- Pull your last 12 months of work. List every project, every shipped feature, every deprecated system.
- For each item, write the prior baseline: how long it used to take, how many people it used to need, what it used to cost.
- Attach the AI leverage: which agent or tool, how many hours saved per week, which hire was deferred.
- Convert to one of the five templates above. Keep the unit (dollars, headcount, weeks) explicit.
- Cross-check against the Klarna table: are you claiming replacement, or claiming leverage? Rewrite anything that reads as replacement.
- Tailor per posting. A vertical fintech role wants GP per head; a horizontal dev-tool role wants revenue per head; an infra role wants cost line removed.
Most companies will follow Block within a year, as Dorsey said. The bullet you write this weekend is the one that clears the rubric they roll out next quarter.
FAQ
What number should I use if my company does not report gross profit per employee?
Use your own team as the unit. Pick a metric your team already tracks (revenue, ARR, deflected tickets, shipped PRs, models in production) and divide by team headcount before and after your AI work. "Grew team-owned ARR from $4M to $9M while headcount stayed at 5" is a defensible, specific claim even if your employer never publishes a company-wide number. Hiring managers care that the ratio is real and auditable in a reference check, not that it matches Block's exact denominator.
Is "used Copilot" or "used Cursor" enough on a 2026 resume?
No. In Refolk's index of AI-fluent senior engineers, tool names are assumed baseline, not differentiators. What separates candidates now is the leverage number attached to the tool: hours saved per week, PRs per week vs. your own prior baseline, headcount deferred, or a vendor contract retired. Name the tool, then name the number. A bullet with only the tool reads like a 2023 resume.
How do I write an AI leverage bullet if I was laid off before I got to ship the AI work?
Anchor on what you scoped, prototyped, or evaluated, and be honest about the stage. "Scoped and prototyped an LLM triage pipeline projected to deflect 30% of tier-1 tickets; layoff halted rollout" is credible and specific. Interviewers know the layoff wave is real (Amazon cut 16,000 in January after 14,000 the prior fall; Meta trimmed its AI unit and plans to cut 10% of Reality Labs), and a clear scoping bullet beats a vague shipped claim every time.
Does the $2M target apply to PMs, analysts, and operators?
Yes, and the framing is often easier. PMs should quantify revenue per PM on their surface area; analysts should quantify decisions supported per analyst or dashboards retired; operators should quantify processes automated and FTE reallocated. The Ahuja quote about a risk model built in a fraction of a quarter is an operator/analyst story as much as an engineering one. Use the same five bullet templates; just swap the unit of output.