a16zEmil Michael: The Department of War Is Moving Faster Than Silicon Valley on AI | The a16z Show
CHAPTERS
- 1:04 – 2:50
Why Emil Michael joined the Department of War: from Silicon Valley to DoD service
Emil Michael describes the personal and professional path that led him from founding and selling a tech company to entering public service. He explains how the White House Fellowship and time spent alongside Secretary of Defense Robert Gates shaped his long-term desire to return and serve.
- •Sold Tellme Networks to Microsoft and sought a break from tech
- •Joined the White House Fellowship and was placed with Secretary of Defense Robert Gates
- •Firsthand exposure to Afghanistan, Iraq, and Pakistan built passion for defense work
- •Returned to government during a moment he viewed as uniquely disruptive and fast-moving
- 2:50 – 4:43
The case for “wartime speed”: reversing post–Cold War slowdown and consolidation
Michael argues the Pentagon adopted a peacetime operating tempo after the Cold War, encouraging industry consolidation and slower innovation. He frames today’s environment—especially China’s historic military buildup—as requiring urgent rebuilding of domestic capacity and faster execution.
- •Post–Cold War “peacetime speed” and the “Last Supper” dynamic encouraged consolidation
- •Defense industrial base shrank to a handful of prime contractors
- •China’s massive buildup exposed U.S. underinvestment and lag
- •Re-domesticating critical inputs (minerals, batteries, supply chain) is central to national strength
- 4:43 – 5:21
Supply chain vulnerability in practice: sanctions, drones, and dependence on China
A concrete example highlights the risk of relying on Chinese-made components. The discussion shows how sanctions can abruptly cut off motors, batteries, and other inputs, creating major readiness and procurement disruptions.
- •Skydio sanctioned by China revealed fragility in component sourcing
- •Defense procurement can be derailed when critical parts become unavailable
- •The issue extends beyond primes to upstream precursors and sub-suppliers
- •Resilience requires rebuilding domestic and allied supply chains
- 5:21 – 6:08
Cutting priorities from 14 to 6: creating focus and momentum as CTO
Upon taking the CTO role, Michael conducts an “inventory” and reduces a long, jargon-heavy set of priority areas to six. He emphasizes that focus and clarity are prerequisites for motivating a large workforce and driving measurable progress.
- •Inherited 14 “critical priority areas,” unchanged for nearly a decade
- •Argues 14 priorities are unmemorable and often expressed as technobabble
- •Condensed to six areas chosen for highest leverage and impact
- •Applied AI elevated to the top and organizational changes followed
- 6:08 – 7:22
Applied AI rollout: moving fast because the department is behind
Michael explains why applied AI became the number-one priority and why speed matters. He reports rapid adoption gains across the department, while distinguishing U.S. intent (augmenting humans) from adversaries’ intent (replacing human decision-making).
- •Moved the Chief Digital and AI Office into his group to accelerate execution
- •Claims the department is significantly behind in AI relative to the broader world
- •Notes adversaries may use AI to reduce reliance on human command and control
- •Adoption jump: from ~80,000 users to ~1.2 million users in about 90 days
- 7:22 – 9:16
Three defense AI buckets: enterprise efficiency, intelligence analysis, and warfighting logistics
The conversation breaks defense AI into three main use-case categories, each with different value propositions. Michael emphasizes the power of using large internal data repositories and AI-enabled throughput gains for analysts, along with logistics and simulation benefits for operations.
- •Enterprise use cases: automate mundane tasks and improve efficiency
- •Intelligence: unlock siloed data and train on decades of satellite imagery
- •Anomaly detection and increased analyst throughput (orders-of-magnitude gains)
- •Warfighting: logistics planning, asset tracking, wargaming, simulations, and operational planning
- 9:16 – 10:56
Commercial AI models in sensitive operations: contract surprises and operational risk
Michael describes a “holy cow” discovery when reviewing prior contracts governing AI use, including extensive restrictions embedded in high-stakes contexts. He raises the risk of vendor lock-in and the possibility of a model shutting off mid-operation due to terms-of-service constraints.
- •Prior contracts contained numerous restrictions on AI use in operational contexts
- •Models were embedded in highly sensitive commands (e.g., CENTCOM, INDOPACOM, SOUTHCOM)
- •Vendor lock-in created single-point-of-failure risk
- •Terms could theoretically interrupt operations, potentially endangering lives
- 10:56 – 14:26
The “Maduro raid” flashpoint: who controls model usage in lawful military operations?
A reported vendor concern about whether its software was used in a major successful operation triggers alarm about external control over U.S. military decision-making. Michael argues the U.S. cannot allow a company’s internal rules to constrain lawful actions authorized through democratic processes.
- •Vendor questioned whether its software supported a major operation, raising trust concerns
- •Michael underscores the stakes of operations and the burden borne by warfighters
- •Frames AI as a foundational substrate akin to the internet—touching everything
- •Rejects the idea that a company’s “constitution” should dictate U.S. command-and-control choices
- 14:26 – 17:01
Democratic oversight and guardrails: laws, civil liberties, and asymmetric adversaries
Michael situates AI policy within the U.S. tradition of contested but structured oversight—Congress sets law and the executive implements it. He warns against unilateral private decisions that affect national security, especially when adversaries may steal models and remove guardrails entirely.
- •Civil liberties debates (post‑9/11, FISA, National Security Act) illustrate governance mechanisms
- •Tech outpacing law is not justification for private actors to override democratic control
- •DoD has longstanding internal directives on autonomy; real-world conflicts accelerate urgency
- •Adversaries may steal U.S. models and remove safeguards, creating asymmetric constraints
- 17:01 – 20:07
Avoiding single-thread dependence: building a multi-vendor frontier AI ecosystem
Michael argues the U.S. needs multiple capable AI partners to prevent renewed lock-in and to ensure continuity of capability. He points to the shifting tech-industry posture since Project Maven, noting that companies once hesitant to work with government can become strong partners.
- •Four frontier AI companies and a small “roster” of elite researchers shape the landscape
- •Goal: ensure multiple engaged vendors so the department is never single-threaded again
- •Project Maven is cited as a lesson in how industry attitudes can evolve
- •Encourages newer companies to avoid repeating past reluctance to support national security
- 20:07 – 22:00
Fixing procurement bureaucracy: simpler requirements, faster cycles, and fixed-price incentives
Michael outlines a push to modernize contracting by reducing bureaucratic friction and replacing bloated requirements with outcome-based goals. He advocates moving away from cost-plus, change-order-driven programs toward firm fixed price structures that reward efficiency and speed.
- •Targets bureaucracy that blocks new entrants from deploying products
- •Critiques thousand-requirement RFPs and unrealistic vendor compliance claims
- •Moves toward outcome-based requirements (range, payload, environment) and industry-led solutions
- •Prefers firm fixed price and risk-sharing over cost-plus and endless development cycles
- 22:00 – 24:36
What startups must deliver: manufacturing scale, real buyers, and faster yes/no decisions
Michael advises startups to build the production and QA muscle needed to scale, which is where primes still have advantages. He also cautions founders to look for real procurement signals—testing and purchasing—not just verbal enthusiasm, and says he’s pushing for faster ‘yes’ and ‘no’ decisions.
- •Startups must cross from prototype to scalable manufacturing and quality systems
- •Borrow selectively from ‘old world’ production expertise rather than reinvent everything
- •Evaluate success by whether the department is testing/buying, not merely praising
- •Cultural shift goal: faster yeses and faster nos so startups can plan and iterate
- 24:36 – 26:48
A call to service: why technologists should consider government work
In closing, Michael shares a personal motivation rooted in immigration, family, and civic responsibility. He argues national strength requires builders who periodically step into public service, and he frames service as an honored profession that should again attract top talent.
- •Personal background: immigrant experience and desire to give back
- •Wants his children to see that the system requires sacrifice and builders
- •Emphasizes respect for warfighters and aligning technology with their needs
- •Calls for renewed prestige of government service, echoing the Manhattan Project era
