May 1, 2026

Tokens that vest

The fourth pillar of engineering compensation arrived in March 2026. It looks like equity. It will end up vesting like equity. The platforms that don't plan for that now will be the next stage of legacy.

When Jensen Huang floated tokens-as-compensation at roughly half of base salary on the GTC stage, the framing landed in HR memos within a week. Tomasz Tunguz wrote about the fourth pillar. Inside three months, “tokenmaxxing” became a noun in performance reviews.

But the framing has a problem. If tokens are compensation, why don’t they vest? Why do they expire monthly? Why are they non-portable? Equity solved those questions four decades ago with cliffs, ratable vesting, double-trigger acceleration, and 409A valuation. Token comp has none of that scaffolding. It’s RSUs without the R.

The Jamaal Glenn critique

The strongest version of the argument against treating tokens as compensation comes from Jamaal Glenn’s piece for Newcomer. Tokens don’t appreciate; rates fall. Tokens don’t vest; you use them or lose them. Tokens don’t transfer; you can’t take them to the next job.

Each of those is true under today’s defaults. None of them are structural. They’re product choices.

What TokenForge proposes

A grant is a row. Total cents allocated. Vesting start, months, cliff. The cron worker rolls accrued forward each month. When you leave, accrued-and-unvested converts to a portable credit: a Stripe-issued voucher redeemable against supported vendors, or a cash equivalent at a published benchmark rate.

That’s it. That’s the whole structural change. Once tokens vest and port, the fourth-pillar framing becomes load-bearing rather than rhetorical.

Why this requires a new system, not a feature

The reason no incumbent will ship this in the next 12 months is that vesting requires payments, vendor partnerships, and legal structure simultaneously. FinOps platforms have the cost data but no compensation primitive. AI observability tools have the usage data but no people primitive. HRIS systems have the comp primitive but no AI plumbing. Building this requires owning all three.

We built TokenForge into the schema on day one — forge_grants is a first-class table next to budgets and ledger. The UI lands in v4. But the data model is locked: when the SEC or local revenue authority asks for a tax-grade audit of who accrued what when, the answer is already there.

Read next: The anti-tokenmaxxing thesis →