Kirkland Is Spending $500M to Build Its Own AI. The Expensive Part Was Never the Code

The Kirkland and Ellis commitment set against where the effort actually sits — $500 million of firm revenue over three to four years starting with $100 million in 2026 on a Palantir-based platform replacing off-the-shelf legal AI tools, beside the staffing of roughly 250 lawyers contributing to the design, more than 180 technologists building it, and over 1,000 lawyers expected to use it — More domain experts shaping it than engineers building it. That ratio is the finding, and it is the opposite of what a cheap-code story would predict.

On Monday I wrote about McKinsey finding that 32% of organisations had decided against buying software because they could build it with agentic coding tools. That is a survey statistic, and survey statistics are easy to nod at.

This week the same decision arrived with a price tag attached.

Kirkland & Ellis, the world's highest-grossing law firm, is committing $500 million of its own revenue over the next three to four years — starting with $100 million in 2026 — to build a custom AI platform on Palantir's technology, rather than rely on off-the-shelf legal AI tools.

1. What was actually committed

The reported shape of it:

The Kirkland & Ellis platform, as reported

One of the largest technology commitments a law firm has made

$500 million of firm revenue, spread over three to four years, beginning with $100 million in 2026.
Built on Palantir, under a multiyear agreement.
Scope covers private-equity fundraising, investor communications, compliance monitoring and document preparation.
Roughly 250 lawyers contributed to the platform's design.
More than 180 technology professionals, inside and outside the firm, will build it.
More than 1,000 lawyers are expected to use it, drawing on the firm's institutional knowledge.

That is not a procurement decision. That is a software company being assembled inside a law firm, with a four-year runway and a headcount most Series B startups would envy.

2. The ratio nobody is quoting

Every write-up leads with the $500 million. The number I would put on the slide is a ratio.

250 lawyers to design it. 180 technologists to build it.

More domain experts than engineers. On a software project.

Sit with that, because it is the opposite of what the prevailing story predicts. If the constraint on building were writing code, and coding agents have made writing code dramatically cheaper, then the engineering side should dominate the effort and the whole thing should be far cheaper than half a billion dollars.

It is not, and it is not. The expensive input is getting what senior lawyers know out of their heads and into a system — the judgement, the precedents, the firm-specific way a deal gets papered. Agents did not make that cheaper. They arguably made it more valuable, because now there is somewhere to put it.

This is exactly the thing I argued survives

In Monday's piece I said what an internal rebuild cannot cheaply reproduce is "accumulated edge cases, the data you hold, the integrations you have already certified, the compliance surface you have already passed — none of it in the spec, all of it in the price." Kirkland's budget is the empirical version of that claim. They are not paying $500 million for code. They are paying it for four years of extracting, structuring and encoding what the firm already knows, plus getting a thousand lawyers to actually use the result.

3. What this says about the 32%

Kirkland makes the McKinsey number concrete at the very top of the market, and it cuts both ways.

It confirms the direction. A buyer with unlimited access to vendors looked at the available legal AI products and decided the important version had to be built. That is a real signal about where the value sits in knowledge-heavy industries, and software founders selling into those industries should take it seriously.

It also prices the alternative honestly. Most of the 32% in that survey are not budgeting $500 million and four years. They are budgeting a couple of engineers and an optimistic quarter. Kirkland is what the decision looks like when someone actually costs it out — and the gap between those two versions is where a lot of build decisions are going to go wrong.

4. If you sell software into a knowledge industry

At the top of the market

The threat is real

Firms with enough scale, capital and proprietary knowledge will build. Kirkland has all three and the highest revenue in its industry. If your entire customer base looks like that, your product needs to be something they cannot assemble from a platform vendor and their own people.

Everywhere else

The threat is smaller

The same decision at a firm one-tenth the size is not one-tenth the cost. Knowledge extraction does not scale down neatly — you still need your senior people in the room. For most buyers the honest build estimate is much worse than the Kirkland ratio implies.

The knowledge, not the workflow

Reposition around

If Kirkland's spend is mostly knowledge capture, then a vendor whose value is workflow is exposed and a vendor whose value is cross-client data, benchmarks or a compliance surface is not. Know which one you are before your buyer asks.

The staffing plan, not the budget

Ask for

When a prospect says they are considering building, the revealing question is not "what have you budgeted." It is "who is writing down what your best people know, and for how long." Most answers stop the conversation.

5. If you are the one tempted to build

The Kirkland numbers are a useful yardstick precisely because they come from a buyer who can afford to get it right.

Domain time, not just engineering

Budget

The 250-lawyer design contribution is the line most internal business cases omit entirely. Those are the most expensive hours in the building, and they are not backfillable. If your plan does not have a number for expert time, your plan is not costed.

Years, not quarters

Assume

Three to four years, with a phased spend. Even with agents writing the code, the schedule is set by knowledge capture and adoption across 1,000+ users — neither of which compresses because generation got faster.

That you are the exception

Check

Building beats buying when the knowledge is genuinely proprietary and genuinely central to how you compete. For a workflow every firm in your sector runs the same way, you are funding a worse version of something already on the market.

There is also a cost floor worth remembering from yesterday's piece on agent token spend: a single refactoring task measured at up to $14.29 in model spend. Build projects of this shape do not just cost salaries and time; they carry a running meter that most business cases have never included.

6. What I would not conclude

This is not proof that buying is dead. It is one firm, in one industry, with unusual economics — a partnership model where the people whose knowledge is being encoded are also the owners funding the encoding. That alignment is rare and it materially changes the calculus.

The $500 million is a commitment, not a result. It is spread over three to four years and it started this year. Nothing has been proven about whether the platform works, gets adopted by those 1,000 lawyers, or returns anything. The interesting checkpoint is 2028, not now.

The staffing ratio is reported, not audited. "Roughly 250 lawyers contributed to the design" covers a lot of ground, from a workshop attendee to someone seconded for a year. I am reading a real signal into it, and it is worth holding that reading loosely.

The honest summary

The largest law firm in the world costed out "build instead of buy" and arrived at $500 million, three to four years, 250 lawyers designing, 180 technologists building, 1,000+ lawyers using. It is the McKinsey statistic with the invoice attached.

The lesson is not that everyone should build. It is that the cost of building was never mostly the code, which means the thing agentic coding tools got cheaper is not the thing that makes these projects expensive. Knowledge capture, integration and adoption are the budget. They were the budget before agents, and they still are.

For founders that cuts in a reassuring direction and an uncomfortable one at the same time. Your buyers can now build the software part more easily than ever. They still cannot cheaply build the part where you know something they do not — and if you cannot name what that is, Kirkland's spreadsheet is not your problem, your positioning is.

Sources: Reuters via Yahoo Finance, "Law firm Kirkland to spend $500 million developing its own AI platform" · Bloomberg Law, "Kirkland's $500 Million AI Gambit Requires a Cast of Hundreds" · Kirkland & Ellis press release on the Palantir partnership · The Global Legal Post · Spend, timeline and staffing figures are as reported; the ratio argument and the cautions in section 6 are mine.

A Third of Your Buyers Have Already Cancelled a Purchase to Build It With AgentsSeptember 1, 2026

One Refactor Cost $14.29. Every File the Agent Read Stayed on the BillSeptember 2, 2026

The Vibe Coding Trap: Why Thousands of AI Startups Are Paying $50K–$500K to Rebuild Their MVPsApril 2026

IdeaToMVP Academy

Want to build with AI — not just read about it?

4-week live cohort for founders. Learn to ship AI agents, scope MVPs, and automate your business — taught by the same team that writes these guides.

Explore the Academy →