Trump's First 100 Days, Tariffs Impact Trade, AI Agents, Amazon Backs Down

Flexport is already seeing the China tariff shock in real shipments: bookings from China to the U.S. are down 60%. CEO Ryan Petersen said prolonged triple-digit tariffs could bankrupt smaller importers whose Chinese supply chains depend on manufacturing capability, not merely cheap labor.

Importers are using bonded warehouses to postpone tariff bills rather than immediately rebuild supply chains. Petersen said firms can hold cargo in U.S., Mexican or Canadian bonded facilities and owe duties only when goods enter U.S. commerce, allowing them to benefit if China tariffs fall before release.

Chamath Palihapitiya disclosed a private dinner with Trump where nuclear war, not tariffs or cultural issues, became the president’s central theme. He said Trump spoke about his uncle teaching him the destructive power of nuclear weapons and presented avoiding war as a deeply held priority.

A new Trump-aligned Washington private club is being built as a curated network for insiders, with roughly ten $500,000 founding memberships. David Sacks said he cannot own it while serving in government but volunteered to be member number one, describing a younger Republican space intended to exclude untrusted reporters and lobbyists.

AI agents are already creating work companies previously considered too expensive to perform. Flexport uses AI for thousands of daily calls matching truckers to loads, while the podcast’s venture firm is automating analysis across 20,000 yearly applications—work it estimated would otherwise consume roughly 5,000 hours.

The bigger AI software opportunity may be labor budgets, not software seats. Box CEO Aaron Levie argued agents can sell outcomes once performed by paralegals or other staff, and estimated most future AI use will involve previously uneconomic work such as contract review, invoice processing and multilingual marketing.

Enterprise AI is still not reliable enough for many regulated workflows without substantial safeguards. Levie said Box’s benchmark—500 documents with 40 requested fields—put the best single-pass model at about 90% accuracy, requiring repeated runs, smaller chunks, tuned prompts and reasoning models for tougher production work.

AI’s near-term advantage is greatest where answers can be mechanically verified. Sacks argued coding and mathematics improve faster because outputs can be compiled or proven, while legal work lacks an equivalent validator; harder domains will therefore require multiple agents or models, additional compute and continued human review.