A Day in 2029: What Will a Patent Attorney Actually Do When AI Does the Drafting?

Rather than trying to predict the future with certainty, we used the format of a fictional working day in 2029 to examine which tasks AI may take over, which responsibilities are likely to remain distinctly human, and how the relationship between attorneys, trainees, clients and AI agents may change.
Here is my take.
My working day begins.
But unlike today, I do not start with a blank Word document, an overflowing inbox and a vague recollection that somewhere in yesterday’s emails there was an invention disclosure I promised to read.
Before I have had my first coffee, several AI agents have already been working.
One has reviewed my calendar. Another has gone through my unread emails and identified the ones that need my attention. For my first client meeting, an agent has reviewed the invention disclosure, checked whether this is a first meeting, a follow-up or a new invention from an existing client, and prepared background notes for me.
This sounds impressive.
It is.
But, so far, no real patent work has actually been done.
Because the most basic question still remains: What is the invention?
I have seen this problem in real life.
The problem behind the invention is explained very clearly, but the invention not.
At this point, a very capable AI drafting agent might be delighted.
Give it enough words and it can produce a beautifully written patent application.
Unfortunately, it may also produce a beautifully written patent application about an invention that nobody has actually understood.
So my job remains to ask the questions.
What changed?
What does the system actually do differently?
What is the technical mechanism?
Which parts are essential?
Which are optional?
What alternatives would achieve the same result?
What can legitimately be generalized?
Only after that discussion does the drafting agent get permission to start.
Perhaps, by 2029, one of the most important skills of the patent attorney will be knowing when to tell the AI: “Not yet” Hold off on drafting the patent application until I have clarified with the client how the invention actually solves the problem and what it does differently from existing systems.
Another example.
A client launches a new composition that is similar to an existing product, but wants to market the new composition for a new indication.
From a purely patent-drafting perspective, there may be an attractive option: draft a broad claim that covers both the new product and the previous product when used for the new indication.
The AI finds that possibility.
Excellent.
Except that the client does not want that.
The broader claim may be legally available. It may even look clever from a patent perspective.
But it conflicts with the client's commercial strategy.
The weakness is not that AI cannot draft broad claims.
Quite the opposite.
The problem is that an AI system, unless properly directed, may optimize for the wrong objective.
The patent attorney has to understand the totality of the client's needs – the legal position, the product, the commercial message, the competitive landscape and sometimes considerations that will never appear in the patent specification itself.
Without that understanding, the AI may produce something technically impressive and commercially unsatisfactory.
To use a Torah analogy, in the book of Exodus, Moses is overwhelmed by the number of disputes he is personally handling.
Yitro gives him what may be one of the earliest recorded management-consulting recommendations:
Delegate- Let capable people handle the routine disputes.
Moses should establish the standards, teach the law and handle the most difficult matters.
That feels remarkably relevant to AI.
The lesson is not:
Delegate everything and disappear.
It is:
Delegate the work that can be delegated, while retaining responsibility for standards, judgment and escalation.
Similarly, a patent attorney in 2029 may therefore be the “boss” of a collection of AI agents doing searches, first drafts, comparisons, support checks, portfolio reviews and procedural tasks.
But delegation is not abdication.
Or, to put it another way: Yitro might have appreciated a good dashboard, but I doubt he would have said, “Let the dashboard decide justice.”

The relationship between senior and junior attorneys will likely change as well.
Today, a trainee may prepare a first draft and send it to a supervising attorney. In 2029, the trainee may arrive at the meeting with an AI agent of their own.
The agent has already searched the prior art, prepared a draft response, suggested claim amendments and produced arguments.
The senior attorney may thus see a much more mature first draft.
And, importantly, the senior attorney may be able to review more of the underlying material.
Today, there are practical limits. A partner cannot necessarily read every document cited in every search, every intermediate draft or every background reference.
An agent can.
It can surface the most relevant prior art, compare versions, identify inconsistencies and show the reasoning trail that led to a particular proposal.
That may make training better, not worse.
But only if the trainee is required to interrogate the agent.
Why did you include this claim element?
What prior art forced this limitation?
Why did you choose this argument?
What alternatives did you consider?
What assumptions are hidden in the draft?
A trainee who simply presses “generate” may learn very little.
A trainee who has to challenge the system's reasoning may learn much faster.
Another big change I foresee may have nothing to do with the quality of AI-generated prose.
It may simply be turnaround time.
Today, I might send a draft back to an attorney for amendment.
Three days later it returns.
By then, I have handled twelve other matters.
I open the file and think: “What was the issue again?”
So I reread the email chain, reopen the prior art, reconstruct the claim strategy and mentally reload the file.
That hidden “re-entry cost” consumes a surprising amount of professional time.
In 2029, the revision may come back in an hour.
Or twenty minutes.
Or three minutes.
I am still mentally inside the problem.
I review it, adjust the instruction and continue.
AI therefore does not merely reduce drafting time. It may reduce context-switching cost – allowing an attorney to handle more matters without repeatedly rebuilding the entire mental model of each file.

if all firms introduce AI tools will they all provide the same quality work?
Suppose the agent proposes something.
A trainee questions it.
An experienced attorney corrects it.
The firm's internal playbook is updated.
The same pattern repeats hundreds or thousands of times.
If that learning remains internal and confidential, the improvement does not necessarily flow back into a generic public model.
That creates an intriguing possibility:
The quality of the AI may increasingly reflect the quality of the humans supervising it.
A firm with excellent lawyers, strong internal review and disciplined feedback may develop an agent that behaves differently from the generic model – or from an agent trained inside an organization where inadequate work is repeatedly approved.
In other words, AI may not eliminate quality differences between professional teams – It may amplify them.

Will we need fewer patent attorneys?
This is the uncomfortable question.
If AI agents are performing searches, preparing drafts, reviewing prior art and revising documents in minutes, then surely one attorney can supervise far more work.
Does that mean we need fewer attorneys?
Maybe.
But that conclusion assumes that the number of matters remains constant.
What if it does not?
Today, many inventions never reach the patent attorney.
Why?
Because invention harvesting takes time.
Someone has to organize the meeting.
The engineers have to prepare something.
The company has to decide whether the idea is worth investigating.
Someone has to review the competitive landscape.
Someone has to follow up.
There is friction at every stage.
AI can lower that friction.
Potential inventions can be identified faster and earlier.
So although each matter may require less attorney time, more matters may enter the system.

So how has my day changed?
Maybe in the beginning, it looked as though the AI agents were doing my job. Scary.
But maybe the conclusion is different and more positive.
AI may have taken over the parts of the job that were easiest to describe.
Search.
Draft.
Summarize.
Compare.
Check.
Revise.
What is still required?
Interviewing.
Understanding.
Challenging.
Interpreting.
Supervising
Connecting legal issues with commercial strategy.
Taking responsibility.
That is, the patent attorney of this scenario is less a document-production machine and more an interviewer, strategist, verifier, orchestrator and supervisor.

So, using another Torah analogy.
When Moshe’s spies return from the promised land with observations.
They all saw the same land, but reached very different conclusion.
Most saw the danger.
But Yehoshua and Caleb saw the opportunity.
Maybe we should too.