d8n.ai
Kazakhstan RU
Uzbekistan RU
Kyrgyzstan
Tajikistan
Azerbaijan AZRU
Explore the demo ↗
AI transformation

Alex Karp's Letter to Shareholders: No Company Will Be Left in the Middle. What It Means for Business Leaders

On November 2, 2026, Palantir CEO Alex Karp published his letter to shareholders for the third quarter. Its central message: every industry will either be remade or set aside, and no company will be left in the middle. The market is splitting into winners and losers, and the gap between them is widening. Below is what the letter actually says and the numbers behind it, why it matters for companies outside the United States, what we have learned from Documentolog’s own AI transformation, and why roughly 70% of success depends not on technology but on people.

What Karp wrote: five theses

Karp’s letters to shareholders are known for their sharp tone. This one is among the toughest — and it is backed by numbers that are hard to dismiss as rhetoric.

One. There will be no middle. Karp writes that every industry — from banks and insurers to hospitals and century-old law firms — will either be remade or set aside. Outcomes will become binary. He describes the middle as a wasteland populated by former leaders who did not listen in time, and below them, companies that once had an edge but never built on it.

Two. Business has overtaken government. For two decades Palantir was known primarily as a contractor to the U.S. government and military. In the third quarter of 2026, its revenue grew 96% year over year to $2.313 billion. U.S. commercial revenue grew 148% to $984 million and, for the first time in the company’s history, exceeded U.S. government revenue of $953 million. Net income for the quarter was $1.128 billion.

Three. Deals have become dramatically faster. According to Karp, a sales cycle that used to take months or years is now compressed into weeks and days. In a single quarter the company closed 212 deals worth at least $1 million, 64 of them worth at least $10 million. This is an important signal: when a large enterprise signs a contract within weeks, the question “do we need AI?” has already been answered on the buyer’s side — the only remaining questions are how, and with whom.

Four. Postponing did not work. Karp writes that some leaders of business and government institutions believed they could delay the future, but change has moved so fast that they have found they “cannot get out of town.” The gradual adoption strategy that worked for previous technology waves may prove too slow.

Five. A model without a system is a liability. The most practical part of the letter is about language models. Karp writes that a model on its own generates limited value. An AI agent that is not connected to the company’s data, people, business units, use cases and safeguards becomes, in his assessment, a virtually unlimited liability. His image is worth remembering: such an agent has a blast radius that tends towards infinity.

Karp draws a conclusion in Palantir’s favor: the only way to connect language models to a large enterprise is through the operating model his company has built. That is to be expected in a letter to shareholders. But the principle itself is broader than any one company, and we will come back to it.

The letter also contains a thought that is easy to miss on first reading. Karp writes that Palantir has no interest in migrating what its customers own and have built into its own systems and accounts. For markets where the question of where corporate data lives — and who can access it — is especially sensitive, this is an important point: a technology partner in AI transformation should strengthen the customer, not take its data.

Why this matters beyond the United States

Business leaders often say: “Let’s wait and see what works for others.” Until recently, that was a rational position. Technology changed every few months, many pilots led nowhere, and the cost of a mistake seemed higher than the cost of waiting.

Karp’s letter records that in the world’s largest market this logic no longer holds. Technology waves reach other regions with a delay, but they always arrive, and the delay is shrinking. When U.S. companies rebuild processes in quarters rather than years, the efficiency gap between those who started and those who wait accumulates faster than before.

For companies in other markets, three practical conclusions follow.

  • The “wait and see” window is closing. Not tomorrow — but we are talking about months, not years. Those months are best spent on preparation: processes, data, people, access rules.
  • Competition is shifting from product to speed of change. A company that moved five processes to AI agents in a year operates at a different cost and speed than one that spent the year in discussions.
  • Data becomes a strategic question. Karp speaks about “sovereign AI” in the context of the United States and its allies. For every other market the implication is simple: the corporate data that AI agents work with should stay within the perimeter and jurisdiction the company itself controls.

The blast radius already exists in every company

Karp’s image seems far from everyday work — until you look at an ordinary office.

A lawyer pastes a contract with a counterparty into a public chatbot to check the risks quickly. An accountant uploads a payroll spreadsheet to find an error. A manager asks for a reply to a key client and attaches the entire correspondence. None of them wants to cause harm; they all just want to work faster. But the company does not know what data has left, where it went, or who can now access it.

This is AI “outside”: it does not know the company’s rules, does not see the approval route, ignores access rights and leaves no trace in the company’s logs. The blast radius already exists — it simply has not been measured. Banning such services is not enough: without an approved tool inside the company, people keep using them on personal devices. The approach that works is to give employees AI inside a system where processes, permissions and accountability are already defined.

What we see from our own experience

Karp’s words resonate with us despite their harshness, because we have walked this path ourselves.

About two and a half years ago, Documentolog approached AI the way business books recommend: we created a separate unit, hired strong specialists and gave them the task of bringing machine learning and AI into the company. A year later, the practical benefit to the business was close to zero. The team was working, but the changes never reached the processes.

Everything changed when the CEO personally took charge of the AI transformation and a cross-functional team — the Documentolog AI core team — was formed inside the company. Today all key processes — records management, contracts, correspondence, customer support, engineering, finance, marketing — are run by our employees together with AI agents. Over the year, the team’s efficiency grew more than fivefold. Customer support headcount went down by 70% — employees moved to complex tasks while the agent took over the routine. In records management the agent removed about 80% of routine work. How this happened is described in detail in “The Six Months When AI Agents Rewrote Our Company”.

From this experience we drew three conclusions that match what Karp writes.

  • Speed really does change in jumps. With each new agent, processes accelerate not by percentages but by multiples, and the effect compounds: the time freed up goes into the next process.
  • The model is the smallest problem. The hardest parts are describing the process, agreeing on access rights, building the knowledge base and deciding which actions the agent takes on its own and which only after human confirmation.
  • The biggest change happens in mindset. Employees start to see their own work differently when they see that the CEO personally uses the results of the AI transformation every day.

Transformation has to be lived through: the 10–20–70 rule

There is no magic pill. You can buy access to a model in a day, but you cannot rebuild how a company works in a day.

Boston Consulting Group puts it as the 10–20–70 rule: in AI adoption, about 10% of the effort goes into algorithms, 20% into technology and data, and 70% into people and processes. In the same publication, BCG notes that the companies doing better are those where the CEO has publicly and consistently declared AI a priority.

Gallup data show the same from another angle. Among workers who strongly agree that their manager champions AI, 33% say AI has transformed how work gets done in their organization. Among the rest, only 4% say so.

Our experience fully confirms this ratio. People need to learn how to assign tasks to agents and check their work. Processes must be rebuilt so that the agent is part of them, not an add-on. You need AI champions — employees who own the agent’s results in their process; why they, rather than the IT department, decide a project’s fate is covered in “Who Actually Implements AI in a Company”. All of this takes time: habits, roles and trust in a new tool do not change by decree.

That is why the earlier a company starts, the better. The time it takes people to adapt cannot be shortened by buying a more powerful model. It can only be started earlier than your competitors.

Agents inside the system, not outside it

Karp is right about the main point: value is created not by the model but by connecting the model to the company’s data, processes and controls. We built d8n — Documentolog’s multi-agent platform — on the same principle, drawing on 19 years of work with document management, approval routes and access rights.

In practice this means a few rules.

  • The agent sees only what is permitted. It works with documents and data within the rights of the employee on whose behalf it acts.
  • Critical actions are confirmed by a human. The agent first prepares a plan — for example, the amount, the recipient and the approval route; a person reviews it, and only then is the action executed.
  • Every step is logged. Time, user, agent, decision and outcome are recorded and available for audit.
  • Data stays within the company’s perimeter. For customers in Kazakhstan, in data centers located in the country; the platform is certified under ST RK ISO/IEC 15408-3-2017 (EAL 4).

What the agent does not solve. Agents make mistakes, sometimes confidently, so they need a company-specific knowledge base, reference answers for verification and regular updates based on employee feedback. An agent will not bring order to a process that does not exist: if nobody in the company knows who approves a contract, the agent will not know either. And an agent does not remove responsibility from the leader — decisions are made by people.

Where a CEO should start

Many leaders want to adopt AI but do not know where to start. Based on our experience, the first 90 days are best structured like this.

  1. Find out what is already happening. Without threatening anyone with punishment, learn which AI services employees use and what data goes into them. Introduce a short rule instead of a ban.
  2. Own the transformation. Declare AI a priority and appoint a person with the time and authority to be accountable for results.
  3. Pick one process. One with a clear owner, a large volume of repetitive work and a measurable outcome: approval time, response time to clients, number of rework cycles. How to choose it is covered in “How to Choose the First Process for AI”.
  4. Define the access perimeter. What the agent reads, what it does on its own, what only after human confirmation, and what is off limits.
  5. Launch in the live system and measure. Not on made-up data, but in a real process with limited permissions and logging. After 90 days, decide on the numbers: scale up, rework, or shut down.

Frequently asked questions

Does Karp’s letter apply to companies outside the United States? Karp writes about the U.S. market, but he describes a general shift: AI is changing the economics of every industry, and buyers are deciding much faster. For other markets the question is not whether this will happen, but when.

Is it enough to buy access to a language model? No. A model on its own generates limited value. The effect appears when the agent is embedded in processes, works with the company’s data under the company’s rules and remains under control. Under BCG’s 10–20–70 rule, technology and data account for 30% of the effort; the other 70% is people and processes.

Why not simply allow employees to use public chatbots? Because the company loses control over its data: contracts, financial spreadsheets and client correspondence end up in external services, and the AI’s actions leave no trace in the company’s logs.

How long does an AI transformation take? The first process can be moved to an agent within a few months. Rebuilding the company takes longer, because it requires changing habits, roles and processes. That is why it makes sense to start with one process — but to start now.

Where to start if the company has no AI yet? With an inventory of what employees already use, with the CEO’s personal involvement, and with one measurable process. The step-by-step plan is in “Where a CEO should start” above.


Sources

  • Alex Karp, Letter to Shareholders, Q3 2026, November 2, 2026 — palantir.com
  • Palantir Reports Q3 2026 Results, press release, Form 8-K, November 2, 2026 — sec.gov
  • BCG, When Companies Struggle to Adopt AI, CEOs Must Step Up, April 24, 2025 — bcg.com
  • Gallup, AI’s Effect on Workplace Culture, 2026 — gallup.com
CEO DIGEST

Read new materials first

Longreads when published and one weekly digest. Email confirmation is required.