An EA I know checks the same list every single morning before she
opens her inbox. Which resolution cases have an IRS response deadline
inside the next thirty days. She built that list by hand, in a
spreadsheet, and updates it herself, every day, because she got burned
once. A case sat quiet for six weeks, the deadline passed, and she spent
a Saturday writing a very uncomfortable letter to a very angry
client.
She does not need a better spreadsheet. She needs something that
checks the list for her and only bothers her when something is actually
wrong. That is the difference between chatting with AI and building a
tool with it, and almost nobody in this profession has made that jump
yet.
Here is the answer up front: Claude Code lets you describe a task in
plain English and get working software back, not just an answer to a
question. You are not learning to program. You are directing something
that already knows how, and the output is a tool that keeps running
after you close the laptop.
Most Tax Pros Are Stuck at “Ask It a Question”
If you use AI at all, you probably use it the way you’d use a very
well-read colleague. You ask something, it answers, you take the answer
and do the work yourself. That is genuinely useful. It is also the
ceiling of what a chat window can do.
A tool is different. A tool runs on its own schedule, checks
something specific, and only surfaces when a human needs to act. The
deadline monitor my EA friend eventually built does not answer
questions. It checks her case list every night at 7pm and sends her
exactly one message: which cases have a deadline inside thirty days, and
which of those have had no IRS contact logged in two weeks. Nothing
else. She reads three lines and knows if tonight is a work night.
That shift, from asking to building, is what Claude Code actually
does for a tax practice. You are not hiring a developer. You are telling
it what “done” looks like, and it writes the thing.
The Three-Question Test Before You Build Anything
Not every task deserves a tool. Before you spend an hour building
something, run it through three questions.
Does this happen every week or more. A one-time task is not worth
automating, no matter how annoying it is.
Does the same input produce the same output every time. If the task
requires a judgment call, a tool cannot do it. If it is genuinely
mechanical, a tool can.
Is the payoff worth the build time. A ten-minute weekly task that
takes two weeks to automate is a bad trade. A ten-minute daily task that
takes an hour to build pays for itself inside three weeks.
Pass all three and it is worth building. Fail any one and leave it
alone.
What Passes This Test in a Tax Practice
Three examples that clear the bar every time.
A deadline monitor. It scans your case list for any IRS response
date, CDP hearing deadline, or offer follow-up window inside a set
number of days and flags it before you find out the hard way.
A document-chase tracker. New client signs the engagement letter, the
tracker logs what documents were requested and when, and it flags anyone
who has gone quiet for more than a set number of days so the follow-up
email actually gets sent instead of living in your head as guilt.
A notice triage log. Every incoming IRS notice gets logged with type,
deadline, and client, sorted so you see the whole pile at a glance
instead of a stack of envelopes on the corner of your desk.
Here is what does not pass the test. Anything involving a filing
position. Anything that happens twice a year. Anything where being wrong
creates real exposure for you or your client. Those stay human, always.
Build the tool for the mechanical checking. Keep your license and your
judgment on the parts that need it.
How the Build Actually Works
You do not open a code editor and start typing syntax. You describe
the outcome in plain language. What should this check, how often, what
counts as a problem worth flagging, and how do you want to hear about
it. Claude Code takes that description and builds the working version.
You look at it, you tell it what is wrong or missing, it fixes it. That
loop runs a handful of times and you have something real.
The mistake most people make on their first build is asking for too
much at once. Start with one narrow thing. The deadline monitor, and
nothing else. Get that running for two weeks. Then add the second thing.
A tool that does one job well beats a tool that tries to do five jobs
badly, every time.
Start It in Draft Mode. Always.
Every tool you build should start in a mode where it tells you what
it found, not one where it acts on its own. It flags. It drafts. It does
not send anything to a client and it never touches an IRS system on its
own. You review, you approve, and only after weeks of watching it get
things right do you loosen the leash at all. In a licensed profession,
that caution is not optional. An automated tool that sends the wrong
thing to a client is not a customer service problem. It is a
professional one.
The Confidentiality Line
This matters more here than it does anywhere else this gets taught.
Do not paste full returns, full transcript sets, or anything with a
Social Security number into a consumer-grade AI tool. Use a business or
enterprise tier that does not train on your data. Redact names and
identifying numbers before anything goes in. Keep a written list of what
tools you use and what you allow them to touch. This belongs in your
firm’s written information security plan, not as an afterthought.
Section 7216 governs disclosure of taxpayer information, and pasting a
client’s data into the wrong tool is a disclosure question, not a
convenience question.
What This Actually Buys You
Read that again: the tool checks it for you every night, whether you
remember to or not, whether you’re at your kid’s game or asleep or three
states away at a conference. That is the whole point. You stop being the
human cron job that has to remember to check the spreadsheet, and you
start being the person who reviews what the tool already flagged.
I get it. I respect it. Building something for the first time feels
like it belongs to a different kind of professional than you. It does
not. You already run a process-driven business. You pull transcripts,
you read the account, you match the taxpayer’s numbers against the
standards, you pick the right path. Building a tool is the same
discipline pointed at your own operation instead of at a case.
I teach this build, live, hands on, at our AI Summit. Day two of that
event is entirely Claude Code, start to finish, so you leave with
something running, not just notes about something you might build
someday. Check taxresolutionacademy.com for the current dates and a free
seat.
Reply and tell me the one repetitive task that eats the most of your
week. I read every one, and there is a decent chance it is the exact
thing worth building first.
I am on a mission to improve the lives of licensed tax professionals
by showing them there are more options in IRS representation than they
were ever taught, so they can raise client quality, grow revenue, cut
the hours, and get their life back. A tool that runs itself while you
are at dinner is a small piece of that. It adds up.
If posts like this one belong in your inbox, drop your email below. One post a week. No fluff.
Here’s to working smarter, not harder!
And a brighter future for your tax practice!
If you want to know more consider joining the Tax Resolution Academy® and earn your Certified Taxpayer Representative™ (CTR™) certification.
I hope this helps.
If you have any questions, please reach out to us.
Have a GREAT day,
With Love,
Dan
Dan Henn, CPA, CTR™
Co-Founder, Tax Resolution Academy®
Managing Member
Tax Pro Academy, LLC
P.S. Want to learn more about the Tax Resolution Academy®, go to https://members.taxresolutionacademy.com.
P.P.S. Check out our CPE classes here! Click the link to Follow our page and you will be notified when we release new classes.
