Thinkshoring: Outsourcing Your Judgment to AI (Don't)

Somewhere on your team, a deliverable is moving toward a client that no human has actually read. The internet has already invented insults for the person behind it, including "slopper" and "secondhand thinker." Name-calling does not help you manage the problem, though, because the problem is not a person. It is a behavior, and a behavior needs a name you can put in a policy. We propose thinkshoring.

What Is Thinkshoring?

Thinkshoring is the practice of "offshoring" your thinking: sending work that requires your judgment to an AI chatbot and shipping whatever comes back, without engagement, without review, and without the ability to defend the result.

The word borrows deliberately from offshoring, because 30 years of outsourcing to low-cost jurisdictions taught business a lesson that transfers almost perfectly. Offshoring worked when companies kept the specification, the quality control, and the accountability in-house. It failed when they shipped the entire function overseas and hoped for the same or better output at a drastically lower cost. AI presents the same bargain on a personal scale, at a faster tempo and a fraction of the cost. You can outsource production. You cannot outsource accountability.

That distinction matters because thinkshoring is not the same thing as using generative AI tools. Delegating to AI can be effective and, for competitive teams, close to mandatory: you brief the tool, you review and refine the output, and you own the result. Thinkshoring is abdication wearing delegation's clothes. The prompt goes in, the output goes out, and at no point does a human form a view. The tell is simple. Ask the sender a follow-up question. A delegator answers it. A thinkshorer asks the chatbot.

How to Spot Thinkshoring on Your Team

Thinkshoring has recognizable field marks:

• The unanswerable memo. The "author" cannot field the first follow-up question, because the "author" never formed the view the memo asserts.
• The reply to an unread thread. The response addresses points nobody raised, which usually means nobody read the thread before prompting.
• Confident nonsense in the details. Dates, figures, or clause references arrive plausible, polished, and…wrong, which is the signature texture of unreviewed AI output.
• Volume without progress. The documents multiply, the decks lengthen, and no decision moves, because output has been mistaken for work.

The receiving end of thinkshoring already has research behind it. BetterUp Labs and the Stanford Social Media Lab coined workslop to describe AI-generated work that looks finished but fails to advance the task. In their survey of 1,150 US desk workers, roughly 40% had received workslop in the prior month, each incident took about 2 hours to resolve, and the invisible tax ran to about $186 per employee per month, which the researchers scaled to more than $9 million a year for a 10,000-person organization. The relational damage compounded the financial cost: roughly half of recipients viewed the sender as less capable, and 42% viewed the sender as less trustworthy.

"Workslop" names the artifact. "Thinkshoring" names the habit that produces it.

The Deeper Costs: Capability and Accountability

The rework cost is the visible part. Two quieter costs compound underneath it.

The first cost is capability. A widely covered 2025 MIT Media Lab study of essay writers found that participants who leaned on an AI assistant engaged less deeply with the task and remembered less of their own work, a pattern the researchers described as accumulating "cognitive debt." A team that thinkshores for a year has not saved a year of thinking. It has skipped a year of training, and it will feel the gap in exactly the moments AI cannot cover.

The second cost is accountability, and courts have already ruled on it without needing the word. In Mata v. Avianca, a 2023 case in the Southern District of New York, 2 lawyers filed a brief citing 6 cases that ChatGPT had invented, then doubled down with fabricated opinion excerpts when the court asked for copies. The court fined the lawyers and their firm $5,000 and ordered them to mail the sanctions opinion to every judge falsely named as an author. Notably, the court did not condemn AI. The opinion acknowledged that using a reliable AI tool for assistance is not inherently improper. The sanction landed on the thinkshoring: signing work that no human had verified.

United States: Mata v. Avianca is the canonical example, and courts have repeatedly cited it since. United Kingdom and Australia: courts and regulators in both jurisdictions have dealt with comparable incidents involving unverified AI citations, and the professional duty runs the same way. Whoever signs (or adopts), owns.

How to Delegate to AI Without Thinkshoring

The fix is not less AI use, or banning it. The fix is running AI tools the way you would run any capable delegate, under a set of working rules:

1. Form a view first. Write your position before you open the chatbot. The draft then has something to be judged against, and you remain the author of the thinking even when the tool authors the prose.
2. Treat output like a first draft from a talented junior. You would never send a junior's memo to a client unread. The same standard applies here, every time, not only when the stakes feel high.
3. No recursive prompting. You wouldn't ask the junior who wrote the first draft to review their own work. Don't ask the chatbot to evaluate the chatbot's own output, so the loop never touches human judgment.
4. Apply the defend-it test. If you cannot explain a line to a colleague, a client, or a judge, that line does not go out under your name. This test catches invented citations and confident nonsense before they catch you.
5. Name the accountable human. Every AI-assisted deliverable needs a single named owner who stands behind it as if they typed every word, because as far as your client, your counterparty, and the court are concerned, they did.

Teams that follow these rules get the leverage without the liability. Teams that skip them get workslop, rework, and, eventually, a very uncomfortable show-cause hearing (or the equivalent in their profession).

Are you building AI into your legal or contracting workflow? Book a meeting with our team and we will design the review layer in from the start.

Where the Line Sits for Legal Work

We are not AI skeptics. We run AI hard inside our own workflows, across high-volume contract review, first-draft generation, and research triage, because our clients move at market speed and their true competitor is anything slow. The system works because a named lawyer forms the view, interrogates the output, and signs. That review layer is the entire difference between thinkshoring and leverage: the production went out, and the judgment stayed home.

The same principle applies inside your company. Your board deck, your investor update, and your customer contract can all be AI-assisted. Each one still needs a human who can defend every line, because each one carries your signature, your representations, and your risk.

The Takeaway

Thinkshoring is outsourcing your judgment rather than your work, and judgment is the single function that cannot leave the building. Use AI liberally (and learn to use it better). Review everything. Keep a named human accountable for every deliverable, and keep your own skills in rotation.

Stay ahead of AI and the law: sign up for our US updates.