Government Modernization

Making the government not only modern and efficient but aspirationally even cutting edge is a curious problem for a startup for a handful of reasons.

The gov from the outside feels woefully antiquated but then efforts like DOGE have failed to find large, low-hanging fruit and anecdotally report being surprised that it’s reasonably well-run. And while DOGE has been run by almost no one that understands the gov’s functioning, it started with the most power and political aircover any organization no less a random vc-backed startup could ever dream of.
It’s a gargantuan market. The gov’s annual spend on IT at $130B / yr, with $75B in civilian IT budget alone. Individual contracts are often $100M+ and can even be up to $100B (for a single contract!!!). However, of this ~80% typically goes to operations and maintenance of existing IT, including legacy systems. So it’s not as though these figures are about true modernization efforts. E.g. Tech Modernization Fund is ~$200M investment/yr.
The customer sucks. 6-18 month sales cycles in a broadly slow-moving bureaucracy. Budgets predefined years in advance that are often tight. Few employees are tech forward and the end users are often are far far worse from early adopter PoV (e.g. seniors, veterans).
However, customer churn would presumably be extremely low once they have integrated the product
Modernization and efficiency initiatives have been explicit goals of every administration of the last twenty years. While agencies across the board have made gradual progress as seen below and with some serious wins, it’s worth noting that the bar for “modernization” remains low, the Government Accountability Office still considers the federal gov’s management of IT “high-risk” and still highlights that several departments remain shockingly out of date to the point of using 50-year-old systems (hence viral clip of NJ governor pleading for COBOL engineers). Even still, the main complaint that top ranking digitization/modernization gov officials report in testimonials that they don’t have a champion high up enough in the gov (e.g. effectively a CTO of America) nor consistent enough funding. And many of the most proudly advertised public wins have been in making website’s UI nicer, rather than more structural, internal efficiency boosts. All is to say, modernization is definitely happening, but slowly.
The appetite for change has never seemed higher and new rules like FedRAMP 20x seek to cut the time to getting the all-important clearance certification from years to possibly weeks

What to build?

Palantir for Federal Government

This is the lowest hanging fruit of a concept. Forward deployed engineers, do high-touch projects that don’t scale at first but productize those efforts over time. But it begs the questions why Palantir hasn’t leaned in heavily beyond its CDC contract and whether it’d move in with force if a startup showed success.

Maybe a better comp to model such a SotA software vendor is actually Epic Systems that focuses on standardized SaaS products and builds very slowly over time in a high-trust, high-downside, technologically conservative industry.

AI Copilot

A couple people trying this.

Use-Case Specific Government AI Models

There is a version of the world in which AI is brought into government with a single model provider (OpenAI) or a single contractor (Palantir/Anduril). This creates a lot of security risks due to data overlap, biases, ability to be manipulated and more. It feels abundantly clear that a single model and a single provider should not be in charge of multiple government functions or areas of focus, thus, could one build a company (or companies) built around training specific domain focused models that reason for specific problems within government use-cases and decision making.

AI-Generated Formal Verification for Software Development and/or Legacy Migrations

See the respective theses for more on formal verification and computable contracts.

While FV has historically been expensive and slow, AI-generated proofs may make FV cheap and scalable. It could be that most enterprise production code or otherwise code of material importance is automatically formally verified in 10 years time. An interesting aspect of this thesis is that according to a community orchestrator we spoke with none of the FV researchers themselves are scaling-pilled, refusing to believe that AI systems will be capable of this work any time soon. So there should a longer runway without competition for a startup that finds researchers that are all in on disproving that disbelief.

You could build a formally verified Devin specifically designed for government development projects to build the highest quality, most rigorously tested automatically generated software possible. Or you could build one focused on automating legacy system migrations with FV-guaranteed safety and correctness.

Computable Contracts to Automate Government and Regulatory Piping

Computable contracts enable the autonomous deterministic execution of language-based contracts. Just like Amazon invested extensively over decades to completely automate the contract of buying arbitrary things on the internet and having it show up at your door with computable contracts, some piping of the government and especially regulation could be automated. You could build a system that:

autonomously translate existing contracts / law into computable contracts
use LLM-based formal verification to prove the system did it correctly automatically and at scale
use that end reward data for reinforcement learning to further hone the system’s translation capabilities

Having both the deterministic execution of computable contracts and formal verification that the implemented computable contract is guaranteed to perform as intended may be the only way to deploy complex fully autonomous agentic systems at scale.

Where to look to build?

“The Government” is so amorphous with such broad surface area that it’s worth breaking it down with some numbers to illustrate a naive picture of possible customers.

Which departments spend the most to estimate current TAM:

What types of technologies is it currently getting spent on:

$17B goes to cloud spending
$23B on big data-related obligations
$27B on cybersecurity
$5B on AI. Roughly half of usage is on science with just ~1/6th on internal management

What departments might be most tech forward?

The age distribution in local governments skews a bit younger than federal (avg age is 41 vs 45), which may make them more slightly willing adopters. With that said, knocking door to door at each local government begging for $100K contracts makes for brutal ramp.

There’s also been interesting movement in procurement, with OTA DOD procurements going vertical

Comparable Companies

What to build?

Palantir for Federal Government

AI Copilot

A couple people trying this.

Use-Case Specific Government AI Models

AI-Generated Formal Verification for Software Development and/or Legacy Migrations

See the respective theses for more on formal verification and computable contracts.

Computable Contracts to Automate Government and Regulatory Piping

autonomously translate existing contracts / law into computable contracts
use LLM-based formal verification to prove the system did it correctly automatically and at scale
use that end reward data for reinforcement learning to further hone the system’s translation capabilities

Where to look to build?

“The Government” is so amorphous with such broad surface area that it’s worth breaking it down with some numbers to illustrate a naive picture of possible customers.

Which departments spend the most to estimate current TAM:

What types of technologies is it currently getting spent on:

$17B goes to cloud spending
$23B on big data-related obligations
$27B on cybersecurity
$5B on AI. Roughly half of usage is on science with just ~1/6th on internal management

What departments might be most tech forward?

The age distribution in local governments skews a bit younger than federal (avg age is 41 vs 45), which may make them more slightly willing adopters. With that said, knocking door to door at each local government begging for $100K contracts makes for brutal ramp.

There’s also been interesting movement in procurement, with OTA DOD procurements going vertical

Government Modernization

What to build?

Where to look to build?

Comparable Companies

Related Reading

Related Theses

Marketplaces Requiring Private Intelligence

Computable Contracts

Cross-Company Agent Layer

Government Modernization

What to build?

Where to look to build?

Comparable Companies

Related Reading

Related Theses

Marketplaces Requiring Private Intelligence

Computable Contracts

Cross-Company Agent Layer