From Enabling to Doing: Who Wins When the Pentagon Buys Outcomes

SaaS sold access. AI sells labor. The defense market is about to find out the difference.

Author’s note: This is the follow-on to From Seats to Sorties, where I argued the Pentagon should buy software the way it buys weapon systems, on outcomes not inputs. That piece covered the why. This one covers the who, and the answer is less obvious than you’d think.

Last week, I made the case that the government should stop buying software by the seat and start buying it by the result. The C-17 PBL proves it can work, the F-35 proves it can fail, and the SaaS bloodbath proves the commercial market is already forcing the same reckoning.

The response I got most was: Sure, in 20-30 years. The Pentagon moves too slow. While I agree funding cycles move slow, this is such fundamental shift happening before our eyes, I don’t think it’ll be more than 3-6 year change. Which is pretty darn fast in defense terms.

The response I got second most often was some version of: OK fine, but if this actually happens, who benefits?

It’s a fair and more interesting question. And “startups win, incumbents lose” is not a good enough answer.

The best model doesn’t win an outcomes-based contract. The best business model does. Right now, most of the defense technology ecosystem is running on business models that are misaligned with paying for results. To see why, you need a framework for what AI actually changes about where value lands. The sharpest one I’ve found comes from Soren Larson.

Software Sold Nouns. AI Sells Verbs.

For two decades, defense software vendors got rich selling nouns. Palantir sold a platform. Analysts used it to do intelligence work. ServiceNow sold a workflow engine. IT teams used it to manage tickets. The software enabled the work, humans did it, and the vendor got paid per seat, per license, per module, regardless of whether the work produced results.

The commercial world ran the same play: Airbnb modularized homes, Uber modularized cars, DoorDash modularized food. The platform sat in between, enabling without doing. The coordination layer captured the surplus.

AI breaks this model because AI modularizes the verb, not the noun. Waymo doesn’t sell a platform to help humans drive. It drives. Zipline doesn’t sell logistics software. It delivers. The human in the middle shrinks or disappears, and the execution itself becomes the product.

Soren calls this the inversion of Christensen’s aggregation theory: when intelligence commoditizes, value migrates from the enabling layer to the doing layer. The entity that bears liability for the outcome captures the surplus. The entity stuck “enabling” gets squeezed. It’s a clean framework, and I think it explains something important about where the defense market is headed, but it requires translating from the commercial context Soren writes in to the specific weirdness of government procurement.

In defense, the vendor that can actually do the intelligence fusion, the targeting support, the cyber response (not just sell the platform somebody else uses) is the one built for an outcomes-based world. The vendor still selling seats to a dashboard is holding a pricing model that’s about to age badly.

You can’t price a verb by counting the nouns.

Context Is the Moat

Soren’s follow-on insight matters even more in defense than it does commercially: the verb’s value is inseparable from real-time operational context. Waymo can drive because it has persistent access to sensor data, maps, traffic, and years of accumulated context. Remove the context, the verb is worthless.

Apply this to defense. A vendor selling “intelligence fusion as an outcome” can only deliver if it has deep access to the data feeds, the mission parameters, the CONOPS, the commander’s priorities, the feedback loops. A SaaS dashboard sitting on top of somebody else’s data lake doesn’t have that context. A deeply embedded platform with persistent access might. An AI agent with long-running memory and tool access across classification levels definitely does.

This is why the C-17 PBL works and the F-35 PBL failed. Boeing and the Air Force share context through a combined program office at Robins: same maintenance data, same flight data, same scoreboard. Lockheed hoarded its proprietary technical data. The arrangement collapsed. You can’t buy outcomes from a vendor that won’t share the ruler.

The implication for the current defense software market: most of it is stuck in the enabling layer, selling platforms, dashboards, and seats. The vendor doesn’t bear liability for mission outcomes and doesn’t own the operational context. It can’t, because its business model is designed to sit in between, not to do the work.

This is about to determine who wins and who loses roughly $130 billion a year in federal IT spending. So let’s get specific.

The Four Archetypes

If outcomes-based software procurement happens (and I believe it will, whether driven by the XaaS pilot, Congressional pressure, or budget math that becomes impossible to ignore) the competitive landscape restructures around four archetypes. Not all of them survive.

The Primes

Lockheed. Raytheon. Boeing. Northrop.

They already do PBL for hardware. They have the cleared workforce, the Hill relationships, the capital. But software is not their core competency, their proprietary stacks resist independent measurement (see: F-35), and their cultural DNA is cost-plus. They will pursue outcomes-based software contracts by acquiring smaller AI companies and slowly digesting them into traditional cost structures.

I’ve watched this movie before. The acquired company’s best engineers leave within 18 months. The product gets absorbed into the prime’s internal IT governance. The thing that made it good disappears.

Verdict: Positioned for hardware PBL, not-well positioned for software outcomes.

The System Integrators

Accenture Federal. Booz Allen. Deloitte. SAIC. Leidos.

This is the most interesting case, and I want to spend real time on it, because the conventional wisdom (that SIs will adapt and thrive) is wrong in a specific and important way.

SIs are the closest thing defense has to outcomes-oriented services today. They scope by mission deliverables. They’ve done percentage-of-outcome pricing. They understand what “buy a result” looks like in contract language. In a world moving toward outcomes, shouldn’t they be natural winners?

No. Because their economic model is headcount arbitrage.

I’ve sat across the table from SI teams on competitive bids. The proposals are impressive: sharp people, deep domain expertise, polished oral presentations. And somewhere in the cost volume, there’s always a staffing matrix with dozens of FTEs, each billed at rates that would make a SaaS company blush. The capabilities they propose are real. The headcount required to deliver them is the part that’s about to become fiction.

An SI prices every engagement by estimating how many cleared bodies it takes to deliver the result, then marks up the labor. The margin is the spread between what they charge per FTE and what they pay per FTE. It doesn’t matter how many dashboards or AI tools they layer on top. The P&L is driven by billable heads. Every engagement is, at its core, a staffing play with a strategy PowerPoint stapled to the front.

AI makes that model untenable. Not because AI is better at managing people, but because it removes the need for most of the people being managed.

Picture the bid. An intelligence fusion contract, outcomes-based. The government wants a defined number of finished intelligence products per week, meeting specified quality criteria, with measurable time-to-decision improvements. An SI bids 50 cleared analysts, 5 team leads, a project manager, and some Palantir licenses. An AI-native startup bids 5 cleared analysts, 3 engineers, and 50 AI agents running on a few hundred thousand dollars of compute per year.

Same outcome. The AI-native team’s bid is 80% cheaper.

The SI cannot match that price without firing 90% of the team, which is 90% of its revenue on the contract. No CEO walks into an earnings call and says, “Great news, we figured out how to deliver the same results with one-tenth the headcount, and our revenue dropped accordingly.” That’s not a strategy, that’s a resignation letter.

So what will SIs actually do? They’ll adapt the pitch: “AI-augmented services.” Same team structure, now with AI assistants. Sounds modern, looks innovative, preserves the billing.

But augmentation preserves headcount. Outcomes-based competition rewards replacement of headcount. The difference between “AI-augmented” and “AI-native” is the difference between a taxi company that gives drivers a GPS and Waymo. One improves the existing model, the other eliminates the labor the existing model depends on.

SIs will fight to keep the transition slow. They’ll keep requirements complex enough to justify large teams. They’ll argue that human oversight requires human headcount, which is true up to a point, but the point is a lot lower than their billing rate assumes. They’ll use their lobbying muscle to ensure outcomes-based contracting, if it comes, still requires enough warm bodies to protect the spreadsheet.

They are the incumbents with the most to lose and the most sophisticated playbook for making sure they don’t lose it.

I want to be fair here: SIs employ tens of thousands of cleared professionals who do important work. A lot of the institutional knowledge about how the defense enterprise actually functions lives inside Booz Allen and SAIC, not inside government. If SIs contracted overnight, the immediate effect wouldn’t be efficiency, it would be chaos. The question isn’t whether SIs add value today, they clearly do. The question is whether their economic model (margin driven by headcount) can survive a world where the headcount required to deliver the same outcome drops by an order of magnitude. I don’t see how it does….unless…the scope and role that the government takes on in the world expands drastically. Hard/soft power projection increases drastically. Then there is more work to be done by all, and the contraction of the SIs is just an expansion of mission.

The Platform Incumbents

Palantir. Microsoft. Oracle.

I covered the pricing problem in the last piece: both are input-based businesses. Palantir’s $10 billion Army EA is seats, not outcomes. Microsoft’s government business is licenses and cloud consumption. The financial physics of being public companies with per-seat economics makes voluntary transition to outcomes-based pricing nearly impossible.

What I didn’t cover is what they’ll do about it. A lot. And the answer is more interesting than simple resistance: they’ll focus on outcomes AND try to become the measurement layer.

If outcomes-based contracting requires independent measurement of vendor performance, whoever controls the measurement infrastructure has enormous power, even if they’re not the primary vendor. Palantir is already positioning AIP as the “operating system” through which all defense AI runs. Microsoft is positioning Azure and Copilot as the baseline infrastructure. If you’re the platform that measures outcomes, you don’t need to deliver them. You become the scoreboard, and nobody plays without you.

This is smarter than resisting the transition. It’s co-opting it. And it should worry the people designing outcomes-based policy, because independent measurement only works if the measurement layer is genuinely independent, not controlled by a vendor with its own competitive interests.

Verdict: Will try to define the ceiling of reform at a level that protects their model, or reposition as infrastructure. Probably succeed in the short term.

The AI-Native Outcomes Startups

This archetype is just emerging, and I want to be honest about my bias: it’s what I’m building. So take what follows with that context. But I also think the logic is hard to argue with, bias or not.

These are companies designed from the ground up to deliver mission outcomes with AI agents, not to sell licenses or bill by the head. The technology and business model was designed to be the outcome: you pay for intelligence products delivered, targets identified, decisions accelerated, threats detected. The vendor figures out the mix of AI agents, human oversight, and compute required to deliver.

The cost structure is different from everything else in the defense market. The marginal cost of an additional AI agent doing analysis is compute: GPU hours, API calls, storage. Not a cleared FTE at $250/hour with benefits, overhead, and a facility clearance that took 18 months to adjudicate. On an outcomes-based bid, these companies can price at levels that SIs and platform incumbents cannot match without self-destructing.

But the policy environment isn’t there yet. And this is a real problem, not a hand-wave. Multi-year procurement authority for software barely exists. Outcomes-based evaluation criteria don’t exist. Independent measurement standards don’t exist. On a traditional best-value evaluation that weights past performance, existing ATOs, and enterprise agreements, AI-native startups lose to incumbents every time. The product works, the economics work, the contracting vehicle doesn’t.

This is the defense tech valley of death, applied to business models instead of technology.

What this looks like in practice: a SOCOM contract for real-time multi-INT fusion, evaluated on outputs. Threat assessments delivered per day, average latency from collection to product, commander satisfaction scores. The AI-native vendor bids a small team of cleared engineers operating a fleet of specialized AI agents, each fine-tuned for different intelligence disciplines, orchestrated through a secure platform, with an independent telemetry layer feeding performance data to both sides. Total cost: a fraction of the incumbent bid. Not because the people are cheaper, but because there are fewer of them, and the agents do the volume work.

That contract doesn’t exist yet. But the economics are already real. The contracting vehicle just has to catch up.

The capital structure matters too. The SaaS era trained VCs to underwrite customer acquisition: sales teams, marketing spend, land-and-expand. Outcomes-based defense contracting requires underwriting capital expenditure (cleared infrastructure, fine-tuned models, measurement systems, operational integration) before the contract pays a dollar. Same economics as the C-17 PBL: Boeing invests in predictive maintenance knowing a ten-year contract recoups it. A startup on a one-year O&M contract can’t make that bet.

The VCs who figure this out, who underwrite cleared AI infrastructure the way they used to underwrite customer acquisition, will back the winners. The ones still looking for the next per-seat SaaS play in defense are funding the last generation’s model. Multi-year procurement authority doesn’t just help the government, it unlocks the capital structure that makes this entire archetype viable.

Where This Goes

The enabling era produced extraordinary companies and extraordinary margins. It also produced 37,000 WinZip licenses for 13,000 employees and $6 billion a year in software spending that nobody can tie to mission results.

The doing era will produce different winners. Companies that own the verb, not just the noun. Companies that bear liability for outcomes instead of invoicing for access. Companies whose cost structure rewards deploying AI agents rather than growing headcount.

The government has a real choice. Path A: drift toward consumption-based pricing through the XaaS pilot, call it reform, and preserve the existing vendor ecosystem. SIs keep their billing structure. Platform incumbents keep their seats. Waste gets marginally better. The alignment problem stays untouched.

Path B: push toward outcomes-based procurement, accept the political difficulty, and let the market reorganize around who actually delivers results.

Path A is easier. Path B is better. If the last 25 years of PBL have taught us anything, it’s that Path B is possible when the right people push hard enough.

I’ll close with a thought I keep coming back to. The enabling era produced real value for the warfighter, not just waste. It also produced an incentive structure that optimizes for vendor revenue rather than mission results. Those two things aren’t contradictory. They’re the natural consequence of a pricing model that never asked the right question.

The doing era starts by asking it. Not “how many seats do you need?” but “what did this produce?” The companies that can answer that clearly, measurably, independently, will define what comes next. The ones that can’t will have to decide whether to adapt or to lobby.

History suggests they’ll try both. The question is which one works longer.