The Agenda Machine

Most technologies taken for granted today have DARPA roots. This is largely invisible — and not by accident. It is the business model.

What DARPA Is — and What It Says About Itself

#

The Defense Advanced Research Projects Agency was founded in 1958, four months after the Sputnik shock. President Eisenhower wanted an agency that prevents and produces technological surprises by adversaries. DARPA’s own mission statement reads to this day: »The DARPA mission is to create and prevent technological surprise for our national security.«

No education. No civilian research. No public infrastructure. National security.

FY2025 budget: $4.37 billion. For comparison: the National Institutes of Health have ~$47 billion, the National Science Foundation ~$9 billion. With a fraction of those budgets, DARPA shapes tech trajectories for the next decade more decisively than both combined — because DARPA does not ask what scientists find interesting. It decides what the Pentagon needs.

The Model: Industrial Policy Without That Name

#

DARPA has around 100 Program Managers. No career civil servants. Engineers, scientists, entrepreneurs — recruited for 3 to 5 years from industry and academia. Each Program Manager defines their own program, answers the so-called Heilmeier Catechism — nine questions: what is the problem, what do you propose, why now — and then distributes funds to external performers. No peer review. No committees. No tenure. After five years the Program Manager is gone.

That sounds like freedom. It is also the removal of oversight.

DARPA uses the Other Transaction Authority (OTA, 10 U.S.C. §4022) for its contracts — a procurement instrument outside the normal Federal Acquisition Regulation regime. Faster than FAR contracts, fewer transparency obligations, less whistleblower protection. The GAO documented reporting gaps in OTA use across the entire DoD in two reports (GAO-19-511, GAO-22-104021). DARPA is one of the Pentagon’s largest OTA users.

Research economists Pierre Azoulay, Joshua Graff Zivin, Danielle Li, and Bhaven Sampat described the model in an NBER study (2018): DARPA is a state intermediary that actively bridges the »valley of death« between research and product — a form of industrial policy that officially is not supposed to exist in the United States. So it runs through defense budgets.

What DARPA Actually Built — and What Is Myth

#

The attributions run well ahead of DARPA’s own self-description. DARPA’s Innovation Timeline is conspicuously sparse. The agency does not claim everything attributed to it. Precision matters:

ARPANET → Internet: DARPA funded the first networking experiments from 1962 under Director J.C.R. Licklider. In 1969, the first host-to-host login UCLA → Stanford. TCP/IP development (Cerf/Kahn) from 1973 under DARPA funding. The internet has its organizational and conceptual origin in ARPA programs.

GPS: TRANSIT was a US Navy program under the leadership of Johns Hopkins Applied Physics Laboratory — ARPA was newly founded in 1958 and participated in funding, but was not the lead. TRANSIT was the documented predecessor of satellite navigation. GPS itself is an Air Force program from the 1970s built on TRANSIT findings. DARPA did not invent GPS — the contribution lies in the early conceptual stage of satellite navigation, not the system itself.

Siri: From 2003 to 2008, DARPA funded the CALO program (Cognitive Assistant that Learns and Organizes) at SRI International with $150 million over five years. 2007: Siri Inc. spin-off. 2010: Apple acquired Siri Inc. 2011: Siri appeared on the iPhone 4S. Direct line DARPA → mass market product in eight years.

Autonomous vehicles: DARPA Grand Challenge 2004 (no winner), 2005 (Stanford »Stanley«, Sebastian Thrun), 2007 Urban Challenge. Thrun moved to Google, built the Self-Driving Car program — today Waymo. The stacks, algorithms, and core teams of today’s dominant autonomy providers have their personnel origin in Grand Challenge teams.

Moderna / mRNA: In October 2013, Moderna received up to $25 million from the ADEPT program, documented in Moderna’s own press release. Focus: Chikungunya and Zika vaccines, building the mRNA platform. The platform later became the foundation for Spikevax, Moderna’s COVID vaccine. The US Department of Defense has since been investigating whether Moderna failed to disclose the DARPA funding in over 126 patent applications — a legal obligation under the Bayh-Dole Act (35 U.S.C. §202). Knowledge Ecology International has documented the case since 2020. Public sanctions: none so far.

The pattern is consistent. DARPA funds. Industry takes over. Patent ownership rights remain with the recipient. Ties to public interests: negligible or absent.

What DARPA Is Building in 2025

#

The next round’s program is public — at least in part.

Brain-Computer Interfaces (N3): DARPA’s program description is unambiguous: »high-performance, bi-directional brain-machine interfaces for able-bodied service members«, non-surgical. Goal: 16 independent channels in 16 cubic millimeters of brain tissue, read and written in 50 milliseconds. Application image per DARPA: soldiers control drones by thought, command control centers without a keyboard, sense intrusions into networks. Performers: Carnegie Mellon, Johns Hopkins APL, Rice University, Battelle.

AI: DARPA is working on »Third Wave AI« — systems that learn from little data and explain their decisions. Includes the AI Cyber Challenge (AIxCC), with OpenAI, Anthropic, Google, and Microsoft as cooperation partners. The Department of Defense is thus funding AI labs that simultaneously sell civilian products.

Synthetic Biology: Program »Living Foundries« — design of biological systems as factories for materials, fuels, and medicines. Performer network includes MIT, Stanford, Zymergen, Ginkgo Bioworks. The civilian application (biopharma, bio-materials) is an explicit goal — the underlying military question is whether biological production becomes the next strategic bottleneck after semiconductors. Whether and to what extent defensive or offensive biological applications are co-researched is classified. FY2025 AI/IT-R&D share overall: $1.41 billion, ~32% of the total budget.

The Spread of the Model

#

What began in 1958 as a defense project is now the blueprint for American statecraft in every strategic domain:

No parliamentary mandate for the expansion. No public referendum. No debate about whether the right model for energy research should be the same as for weapons development. ARPA-E and ARPA-H replicate Program Managers, OTA contracts, and absent peer review — in fields where the Azoulay-NBER study warns the ARPA model can structurally fail: fast tempo and absent peer review work when a monopsonistic buyer (the military) clearly defines and reliably procures the product. Without that anchor, the risk is that ambitious funding produces expensive individual projects without systemic effect.

The Question Rarely Asked

#

DARPA defenders say: without DARPA no internet, no GPS, no Siri. That is accurate as a fact and misleading as an argument.

The question is not whether DARPA has produced technology. The question is who decided which technologies come into being — and whose interests that reflects. 100 unelected Program Managers with short terms, operating outside normal procurement law, define strategic tech trajectories. Their selection process is not public. The list of rejected program ideas is not public. The revolving door DARPA → venture capital → defense primes is empirically documented, but not systematically studied. Regina Dugan, DARPA director 2009–2012, moved to Google ATAP, then to Facebook Building 8 — now part of Meta Reality Labs. Anduril Industries, founded by Palmer Luckey (Oculus), became a fast-growing defense prime challenger to Lockheed and Raytheon, built on a DARPA alumni network and Founders Fund capital. The pattern: DARPA funds foundational work and talent, VC scales, defense prime or tech conglomerate absorbs.

Sharon Weinberger showed in The Imagineers of War (Knopf, 2017): DARPA was from the start also an instrument for counterinsurgency social science — Project AGILE in Vietnam, the Strategic Hamlet Program. The »purely technical innovator« is a narrative of hindsight, not historical fact.

Annie Jacobsen’s thesis in The Pentagon’s Brain (2015) is sharper: »DARPA creates, DARPA dominates, and when sent to the battlefield, DARPA destroys.« Civilian applications are legitimacy procurement for a military program, not an equal goal.

Neither is wrong. And the internet exists regardless.

That is precisely the problem with DARPA: the products are real, the sender remains invisible, and the question of whose agenda was produced only arises when the product is already in the pocket.

Sources: DARPA.mil (Mission Statement, Innovation Timeline, N3 program description, OTA documentation); DoD Comptroller DARPA FY2025 Master Justification Book; Moderna press release October 2013 (DARPA funding $25M); Knowledge Ecology International (Moderna/Bayh-Dole); Azoulay/Fuchs/Goldstein/Kearney, »Funding Breakthrough Research: Promises and Challenges of the ARPA Model«, NBER Working Paper 24674 (2018); Bonvillian/Van Atta, »DARPA and its ARPA-E and IARPA clones«, Industrial and Corporate Change 2018; GAO-19-511, GAO-22-104021 (DoD OTA use); Sharon Weinberger, The Imagineers of War, Knopf 2017; Annie Jacobsen, The Pentagon’s Brain, Little Brown 2015.