Featuring
Gavin Baker joins Jason and Friedberg for a tour of the debt squeeze and the enormous AI prize before Sacks and Bo Hines report from Washington on the GENIUS Act. The sharpest exchanges are smaller than the episode's headline claims: Jason gets checked on compounding AI productivity, then refuses to let Tether's old reserve record disappear into a victory lap. Gavin has the cleanest episode—bullish without leaving arithmetic unattended.
Spice rack
Did Tether's record justify Jason's skepticism despite its growing Treasury holdings?
Original point: Tether had a messy history, and a U.S. framework should force it to prove liquid backing and clean up before serving American customers.
What everyone argued
Jason Calacanis
Jason called Tether's past 'sordid' and kept returning to the possibility that opaque or illiquid reserves could fail in a run. He welcomed the GENIUS Act precisely because access to the U.S. market would require clearer reserve rules and examinations.
David Sacks
Sacks disputed Jason's characterization, argued that offshore operation reflected the prior U.S. regulatory climate, and said he believed current providers had the money. He emphasized that GENIUS would add reserve backing, disclosures, examinations, and a legal route into the U.S. market.
Bo Hines
Bo backed Sacks by stressing Tether's large Treasury purchases and arguing that bringing every provider inside the same U.S. regime would give the country control and make the system safer.
Winner circle
Jason wins the narrow question. The CFTC and New York records make continued skepticism rational, even though his rhetoric outruns the evidence when he speaks of many government bans. Sacks and Bo make a strong case for the new framework and Tether's growing Treasury role, but those points mitigate the present risk; they do not rewrite the past Jason was contesting.
Commentary
Jason Calacanis
Assumptions and fact checks
Past reserve misconduct remains relevant when assessing present redemption risk.
Why it mattersA history of misleading reserve claims directly affects how much weight to place on voluntary attestations. Current asset growth and new legal duties matter too, so history should inform rather than permanently settle the present assessment.
Tether had previously misrepresented the assets backing USDT and lacked a clean audit record.
CheckThe CFTC found that Tether held sufficient fiat reserves on only 27.6% of sampled days from 2016 through 2018, used non-fiat assets and third parties, commingled funds, and falsely claimed routine professional audits. New York separately imposed an $18.5 million settlement and ended trading with New Yorkers.
Tether had been banned many times by governments.
CheckThe record cited here supports a specific New York prohibition and federal enforcement penalties, not the broader claim of repeated government bans. Exchange delistings or market-access limits under separate regulatory regimes are not automatically government bans of Tether itself.
David Sacks
Sacks is persuasive on why legislation is better than policy-by-administration, but he conflates three questions: why Tether went offshore, whether its old representations were sound, and whether the new law improves future safety. Winning the third does not erase losing the second.
Assumptions and fact checks
Tether's offshore structure was mainly a rational response to hostile U.S. regulation rather than evidence of unresolved governance risk.
Why it mattersRegulatory uncertainty plausibly encouraged offshore activity, but it does not explain away the specific reserve misrepresentations, undocumented arrangements, or commingling found by regulators.
GENIUS safeguards will give consumers enough confidence to expand stablecoin adoption without creating a bank-run-style vulnerability.
Why it mattersOne-to-one liquid reserves, redemption rules, supervision, and disclosure should reduce risk materially. Execution, foreign-issuer comparability, concentration, custody, and run dynamics remain important.
Under the GENIUS Act, every stablecoin must be backed by a dollar in a U.S. bank account and every issuer is subject to quarterly audits.
CheckThe law permits a broader set of liquid reserves, including short-dated Treasuries, qualifying repos, and government money-market funds. It requires monthly reserve disclosure and accounting-firm examination; annual audited financial statements apply to issuers above the statutory size threshold, not quarterly audits of every issuer.
Bo Hines
Bo supplies useful scale but leans too hard on a league-table ranking. The better argument is that new law can make reserve composition, redemption, and supervision consistently testable; size alone is not a receipt.
Assumptions and fact checks
Large Treasury purchases materially answer concerns about Tether's reserve quality.
Why it mattersTreasuries are high-quality liquid assets and the scale is relevant, but gross holdings must be compared with liabilities and verified under consistent reporting and custody controls.
Tether would be the fourth-largest purchaser of U.S. Treasuries in 2025.
CheckTether's own full-year 2025 report says its $28.2 billion of purchases made it the seventh-largest buyer when compared with countries. The episode did not identify a narrower period or alternative ranking that would substantiate fourth place.
Was AI already making developers and businesses dramatically more productive every month?
Original point: Every business could become about 10% more efficient each month, with developers already improving 5% to 10% monthly and potentially doubling annual productivity.
What everyone argued
Jason Calacanis
Jason treated AI adoption as a compounding productivity curve: general-purpose business work might improve about 10% per month, while developers were, in his telling, already shipping 5% to 10% more each month. He used that curve to support a trillion-dollar annual AI revenue pool and a roughly $10 trillion market-cap prize.
Gavin Baker
Gavin called 10% monthly business gains aggressive and stressed that technologies diffuse more slowly than their capability curves. He conceded that software was the clearest early productivity case, but said an actual doubling would amount to an economic revolution rather than a routine extrapolation.
Winner circle
Gavin wins. He accepts that coding is an early success while correctly separating real progress from Jason's unsupported monthly exponent. The later evidence shows meaningful but uneven gains and no demonstrated annual doubling, so the cautious diffusion model better fits what happened.
Commentary
Jason Calacanis
Jason's instinct that AI value will be enormous may age well; his stopwatch does not. The key error is turning a directional trend into a precise compounded rate without measuring which work improved, whether output quality held, or which non-AI bottleneck became binding next.
Assumptions and fact checks
Productivity gains from successive AI releases compound cleanly enough to double business efficiency in roughly a year.
Why it mattersAdoption, task suitability, review costs, bottlenecks, and diminishing returns break simple monthly compounding. By 2026, measured gains had improved but remained far below a demonstrated economy-wide doubling.
One billion workers would rationally pay about $1,000 per year for AI, creating a trillion-dollar revenue market.
Why it mattersThe arithmetic works if the adoption and price assumptions hold, but the episode supplies neither demand evidence nor an adjustment for employer purchasing, price competition, unequal income, or free alternatives.
Experienced developers were already becoming 5% to 10% more productive every month with AI in early 2025.
CheckThe best contemporaneous randomized field study found the opposite in its measured population: 16 experienced open-source developers took 19% longer with early-2025 AI tools across 246 tasks. That study does not represent every developer, but it directly contradicts presenting a large monthly gain as an established average.
Gavin Baker
Gavin does the useful thing in a hype cycle: he grants the strongest observed case and still refuses the exponent. That keeps his caution tied to mechanisms—diffusion and workflow change—rather than generic pessimism.
Assumptions and fact checks
Software would show broad economic productivity effects before most other sectors.
Why it mattersSoftware offers digital inputs, rapid feedback, and cheap deployment, making it unusually compatible with current agents. The later METR evidence shows real gains in some coding settings even though their magnitude remains uncertain.
Short-run diffusion will remain materially slower than frontier-model capability improvement.
Why it mattersOrganizations must integrate tools, redesign workflows, manage quality and security, and train users. Those complements make deployment slower than a benchmark release cycle.
Doubling productivity would be an economic revolution rather than an ordinary incremental improvement.
CheckA sustained doubling of output per unit of labor would dwarf measured late-2025 coding-agent gains and the overall acceleration reported by AI research organizations through spring 2026.
Jason wins the substance by keeping the liability-reserve mechanism in view, but 'sordid' and 'banned so many times' are less useful than naming the CFTC findings. His case becomes stronger, not weaker, when stripped of the fog machine.