Featuring
Brad Gerstner joined the besties for a sprint through AI's SaaS repricing, agent security, and Kevin Warsh's path to the Fed. The sharpest exchanges came when Sacks separated software survival from value capture and Jason applied an opposing-party stress test to Fed independence. Sacks had the strongest episode, while Brad brought the market mechanics and a well-earned Trump Accounts victory lap.
Spice rack
Will AI agents kill SaaS incumbents or move software's profit pool?
Original point: AI creates a headwind because customers may stop believing today's SaaS leaders will remain strong, and cross-application agents could compress per-seat spending while replacing chunks of human work.
What everyone argued
Jason Calacanis
Jason argued from his firm's early agent experiments: cross-tool agents were already taking over work, built-in copilots were less useful, and open or bespoke agents could push SaaS spend from roughly ten percent of an employee's salary toward one percent. He predicted incumbents would need much leaner cost bases and substantially better products.
David Sacks
Sacks rejected the idea that established SaaS systems such as Salesforce would simply be replaced by freshly generated code. His narrower thesis was that some weak or overpriced apps would be removed while durable systems of record survived beneath a cross-tool AI workspace that captured the new value.
David Friedberg
Friedberg argued that software is moving from helping people do work, to completing work, to doing work humans could not perform. He expected outcome-based pricing and software's expansion into services to make the total software market four to ten times larger in five years, though unevenly distributed.
Brad Gerstner
Brad said serious investors were not claiming all software would disappear; they were repricing uncertainty about future cash flows and a profit pool moving toward agents. He argued that data platforms benefiting from AI could reaccelerate while thin application layers faced permanently lower multiples.
Winner circle
Sacks wins by splitting an overbroad question into the two mechanisms that matter: replacement and value capture. Current Salesforce results fit his model of a durable system of record adding an agent layer, while Jason's broad spend-collapse forecast and Friedberg's four-to-ten-times expansion remain untested. Brad's valuation framework was strong, but Sacks best answered what actually survives and where the pressure moves.
Commentary
Jason Calacanis
Assumptions and fact checks
Cross-application agents will usually be owned by open-source projects rather than incumbent SaaS vendors.
Why it mattersOpen agents have flexibility, but enterprise buyers also value identity controls, audit logs, support, permissions, and liability. The market is too young to know which ownership model wins.
Product managers, designers, and developers will routinely consolidate into one AI-amplified role.
Why it mattersAI clearly expands individual scope, but production software still requires specialized judgment, accountability, and review. Consolidation will vary by product risk and organizational scale.
David Sacks
Sacks avoided both denial and apocalypse. His distinction between a product surviving and its future value migrating elsewhere directly answered the central question and handled the strongest opposing case.
Assumptions and fact checks
Enterprises will retain mature systems of record because generated replacements cannot quickly match decades of testing and operational knowledge.
Why it mattersMigration risk, integrations, permissions, compliance, and accumulated edge cases create real switching costs. AI can lower replacement cost without erasing those constraints.
The main AI profit pool will sit above existing SaaS databases in a cross-application workspace.
Why it mattersThat is a plausible architecture, but incumbents can bundle agents, control data access, or acquire successful agent vendors. Who captures the margin remains unsettled.
Claude Cowork-style agents can work across multiple desktop applications and connected data sources rather than remaining inside one SaaS product.
CheckAnthropic documents Cowork plugins and connectors that operate across Microsoft applications and third-party data providers, supporting Sacks's description of a cross-tool agent layer.
David Friedberg
Friedberg expanded the frame beyond defending today's SaaS vendors, which was valuable. The argument needed a clearer bridge from greater technical capability to durable software margins rather than assuming that all newly created value accrues to software companies.
Assumptions and fact checks
A large share of SaaS will shift from per-seat subscriptions to pricing based on completed business outcomes.
Why it mattersAs agents perform discrete work, usage and outcome measures become more natural than seats. Contracts will still need to allocate quality, causation, and failure risk.
The aggregate value of software companies will rise four to ten times within five years.
Why it mattersThe direction is possible, but the magnitude and deadline were asserted without evidence and depend on adoption, competition, pricing power, and how value splits among models, chips, services, and applications.
Brad Gerstner
Brad gave the best investor explanation for the selloff, but his conclusion was more persuasive as a risk framework than as a demonstrated causal attribution. A broader, consistently defined dataset would have made the valuation claim much stronger.
Assumptions and fact checks
AI will permanently reduce the terminal multiples of many application-software companies even when their near-term results remain healthy.
Why it mattersAI increases uncertainty and competition, but a permanent multiple reset also depends on interest rates, growth, margins, switching costs, and whether incumbents monetize agents themselves.
Salesforce's revenue was still increasing even as investors questioned the durability of future SaaS cash flows.
CheckSalesforce later reported first-quarter fiscal 2027 revenue of $11.1 billion, up 13% year over year, and subscription and support revenue up 14%. The result supports Brad's narrower point that current revenue growth had not disappeared.
Should a president pressure the Fed when its chair appears too slow?
Original point: If a Fed chair is too slow to cut rates, harms the economy, and has dug in against the executive branch, the system needs a way to respond.
What everyone argued
Jason Calacanis
Jason warned that weakening Fed independence for a favored president creates the same power for a future ideological opponent. He supported Warsh's selection but opposed presidential control of rates and called for the Powell investigation to end.
David Sacks
Sacks argued that independence cannot become immunity from accountability when a chair is late to respond and harms employment or growth. He also suggested the Fed's lagging data, especially shelter data, could explain slow policy and should be modernized with larger private-sector datasets.
Winner circle
Jason wins the institutional question. His symmetry test exposed the missing guardrail in Sacks's argument, and broad evidence links central-bank independence with lower inflation and stronger credibility. Sacks correctly identified accountability and data quality as problems, but his later modernization proposal solves them more safely than presidential pressure would.
Commentary
Jason Calacanis
Jason's opponent-party test was the most disciplined argument in the exchange. He would have improved it by distinguishing lawful appointment and oversight from coercion aimed at a particular rate decision.
Assumptions and fact checks
Presidential leverage over monetary policy will eventually be used by an administration Jason strongly opposes.
Why it mattersInstitutional powers persist across elections, so evaluating them under an opposing administration is a sound design test even though Jason's named hypothetical ticket was rhetorical.
The investigation of Powell functioned as political lawfare rather than a neutral enforcement action.
Why it mattersThe timing and pressure create a legitimate concern, but motive is not independently provable from the episode evidence and should not be stated as settled fact.
Kevin Warsh would replace Jerome Powell as Federal Reserve chair if confirmed.
CheckWarsh was confirmed by the Senate in May 2026 and took office as chair on May 22, replacing Powell through the established nomination and confirmation process.
Greater operational independence for central banks is associated with better inflation outcomes.
CheckIMF analysis finds higher central-bank independence associated with lower inflation, and the Federal Reserve says international experience links independence from political influence to better economic outcomes.
David Sacks
Sacks's modernization proposal was much stronger than the implied case for presidential intervention. Better inputs and transparent accountability address slow policy without giving elected officials direct control over the rate path.
Assumptions and fact checks
Powell delayed rate cuts because he was dug in or hostile toward the executive branch.
Why it mattersThis assigns motive without evidence. A policy disagreement can arise from the Fed's inflation and employment assessment without personal animus.
Private, real-time rental datasets are categorically better than CPI shelter data for monetary policy.
Why it mattersPrivate new-lease data are timelier, but CPI shelter intentionally measures the rent paid across the housing stock, including continuing tenants. The datasets answer different questions and are strongest when used together.
U.S. employers announced roughly 100,000 January 2026 job cuts, with 30,000 tied to UPS, 16,000 to Amazon, and 7% attributed to AI.
CheckChallenger reported 108,435 announced cuts: 30,000 at UPS after its Amazon split, 16,000 at Amazon, and 7,624 cuts, or 7%, citing AI.
The Fed measures housing inflation by surveying about 8,000 households for their rent.
CheckThe figure is close, but the agency and unit are wrong: BLS, not the Fed, collects about 8,000 rental-unit quotes each month for CPI shelter. The sample is split into six panels, with each unit generally repriced every six months, and it intentionally includes continuing rents rather than only new leases.
Jason supplied the sharpest user-level example, but treated a thrilling internal prototype as a reliable market forecast. He would have been stronger with renewal, seat-reduction, and total-cost data from multiple companies rather than a single firm's experience.