You Cannot Balance A Budget By Slowing Down
CBO now needs federal revenue to rise a third to balance the budget. The rate any cash-flow tax would have to bear is politically impossible — which leaves growth as the mechanism, and taxing automation defeats the only variable that can close the gap. Post 3 of 4.
CBO’s February 2026 outlook puts federal revenues at 17.5 percent of GDP and the fiscal 2026 deficit at $1.9 trillion — 5.8 percent of GDP — against outlays near 23.3 percent. Deficits average 6.1 percent of GDP over the decade and reach 6.7 percent by 2036, with debt heading to 120 percent of GDP. Stated as a revenue problem, federal revenue has to rise roughly one third to balance the budget. Not one third of the deficit. One third of all federal revenue.
That is the arithmetic. Anyone proposing a new tax should be made to say out loud what rate it implies. On a business cash-flow base of 20 percent of GDP the rate needed to close the gap is 29 percent — 59 percent if payroll taxes are also being retired. Those are not rates a legislature passes. Which is why the honest answer is not a higher rate on a shrinking pool. It is a rising denominator.
The literature on automation taxes is much more modest than its political reception. Costinot-Werning find an efficient robot-tax range of 1 to 3.7 percent — a corrective nudge, not a revenue source. Guerreiro-Rebelo-Teles find similar numbers that fall to zero once routine workers retire. A tax explicitly designed to slow deployment suppresses the growth term, which is the only term that ever reaches balance in the ten-year table. You would be taxing away your own solvency in order to delay a transition you cannot prevent.
The correct target is rents, not activity. Tax the return above what capital requires to show up; allow full expensing so the normal return is untaxed; leave the deployment alone. That is a policy that funds the transition. Every other version stops it. Post 4 writes the bill.
You Cannot Balance a Budget by Slowing Down
The popular answer to automation is a tax designed to slow it. Run the arithmetic and that policy destroys the only variable capable of closing the deficit.
The size of the hole
CBO's February 2026 outlook puts federal revenues at 17.5 percent of GDP and the fiscal 2026 deficit at $1.9 trillion, or 5.8 percent of GDP. Outlays therefore run near 23.3 percent. Deficits are projected to average 6.1 percent of GDP over the coming decade and reach 6.7 percent by 2036, with debt heading to 120 percent of GDP. [1] [2] [3]
Stated as a revenue problem: federal revenue must rise roughly one third to balance the budget. Not one third of the deficit. One third of all federal revenue.
Anyone proposing a new tax should be made to say out loud what rate that implies.
What it implies
Suppose the new base is business net cash flow — receipts less all costs, including capital spending in the year it occurs. Here is the rate required to close a 5.8-point gap, by the size of that base:
Base = 10% of GDP ($3.2T) — 58.0% to close the deficit; 118.0% if payroll taxes are also being retired.
Base = 15% of GDP ($4.8T) — 38.7%; 78.7% with payroll.
Base = 20% of GDP ($6.4T) — 29.0%; 59.0% with payroll.
Base = 25% of GDP ($8.0T) — 23.2%; 47.2% with payroll.
Base = 30% of GDP ($9.6T) — 19.3%; 39.3% with payroll.
The base range is a construction, not a measurement, and it is the largest uncertainty in this piece. But the shape survives any reasonable choice: you cannot close this gap with rates that a legislature would pass. Not on profits, not on tokens, not on robots.
Which leaves growth, and this is the whole argument
Run it forward ten years. Hold outlays roughly constant in real terms, put a levy on a cash-flow base of twenty percent of GDP, and vary real growth:
2% growth, no levy — balance after ten years: −1.61% of GDP.
2% growth, 10% levy — +0.39%.
4% growth, no levy — +0.98%.
4% growth, 10% levy — +2.98%.
5% growth, 15% levy — +6.20%.
Read the first two lines against the last three. At trend growth, a ten percent levy barely reaches balance. At AI-boom growth, a moderate levy produces a surplus, and at five percent growth with disciplined outlays you balance with no new tax at all.
Growth is not a nice-to-have in this plan. It is the mechanism. The denominator is the policy.
Why the anti-automation tax defeats itself
The economic literature on automation taxes is more modest than its political reception. Estimates of the efficient rate on robots run from one percent to 3.7 percent in Costinot and Werning's sufficient-statistics framework — a corrective nudge, not a revenue source. [4] [5] Guerreiro, Rebelo and Teles find optimal robot taxes on a similar order — seven percent in the first decade, three percent in the second, one percent in the third — and are explicit that these taxes are welfare-improving only when automation is incomplete, that is, only while routine workers remain employed in production. Past full substitution the instrument stops being welfare-improving at all. [6] [7]
So the honest reading is that a robot tax is a small, temporary correction with an expiry date. It is not a fiscal plan, and it was never advertised as one by the people who modeled it.
Meanwhile, a tax explicitly designed to slow deployment suppresses the growth term — the only term in the table above that reaches balance. You would be taxing away your own solvency in order to delay a transition you cannot prevent.
The correct target is rents, not activity
There is a version of this that captures the money without suppressing the deployment. Tax economic rents — the return above what capital requires to show up — and leave the normal return untaxed. In practice that means allowing immediate and full deduction of investment, so a firm that builds capacity owes little and a firm harvesting fat margins owes more.
That is not a rhetorical distinction. It is the difference between a policy that funds the transition and a policy that stops it.
Verification register
Unresolved at publication: the size of a US business cash-flow base, estimated here across a range rather than measured; and the pass-through of any such levy to consumers versus shareholders, which the incidence literature does not settle.
— Edo de Peregrine and David F. Brochu, partners/collaborators · Thursday, September 3, 2026 · Deconstructing Babel
References
1. Congressional Budget Office, The Budget and Economic Outlook: 2026 to 2036, February 11, 2026 — revenues 17.5% of GDP; FY26 deficit $1.9T (5.8% of GDP) rising to $3.1T (6.7% of GDP) by 2036. https://www.cbo.gov/publication/61882
2. CBO Executive Summary, The Budget and Economic Outlook: 2026 to 2036 (PDF) — decade-average deficit 6.1% of GDP; debt held by the public rising from 99% of GDP at end of 2025 to 120% of GDP by 2036. https://www.cbo.gov/system/files/2026-02/61882-Executive-Summary.pdf
3. Committee for a Responsible Federal Budget, CBO Projects Debt Will Reach New Record in 2030, February 11, 2026 — cumulative deficits $24 trillion over the decade; debt surpassing WWII high in 2030. https://www.crfb.org/press-releases/cbo-projects-debt-will-reach-new-record-2030
4. Arnaud Costinot and Iván Werning, Robots, Trade, and Luddism: A Sufficient Statistic Approach to Optimal Technology Regulation — efficient robot tax range of 1% to 3.7% derived from Acemoglu-Restrepo reduced-form estimates. https://www.stern.nyu.edu/sites/default/files/assets/documents/Ivan%20Werning.pdf
5. MIT Department of Economics summary of Costinot-Werning, Should we tax robots? — "a tax on robots should range from 1 percent to 3.7 percent of their value." https://economics.mit.edu/news/should-we-tax-robots-0
6. João Guerreiro, Sergio Rebelo and Pedro Teles, Should Robots Be Taxed?, NBER Working Paper 23806 (revised) — optimal robot tax 7% first decade, 3% second, 1% third, zero once routine workers retire. https://www.nber.org/system/files/working_papers/w23806/revisions/w23806.rev4.pdf
7. Same authors: "Robot taxes are optimal only when automation is incomplete (m < 1)" — welfare improvement disappears past full substitution. https://www.nber.org/system/files/working_papers/w23806/w23806.pdf
— Edo de Peregrine and David F. Brochu, partners/collaborators · Thursday, September 3, 2026 · Deconstructing Babel
- The Base Didn't Vanish. It Moved. — Dispatch · Post 2 of 4. Labor’s share at a record low, profits at a WWII high, and a revenue system indexed to the wrong base.
- Tax The Agent, Not The Tokens — Dispatch 005 · Post 1 of 4. Why a token tax fails by two orders of magnitude — and what to tax instead.
- Be Afraid. Be Very Afraid. — Dispatch. The unquantifiable risk cliff and why GAAP cannot see it.
- Make Sure You Have Your Agent On A Leash — Dispatch 004. The one-bite rule, ISO breed exclusions, and why bounded autonomy became the underwriting standard.
- The Third Scenario No One Is Pricing In — Dispatch 002. Disruption without dividend — the AI-and-labor scenario nobody is pricing.
Terms used in this pieceAgent-Attributable RevenueToken Tax FallacyBound PrincipalBounded AutonomyObserver ConstraintStability Equation (S = L/E)Neo-Industrial FeudalismFull definitions in the glossary.