It is 8:47 AM and I am already three tabs deep into a corporate structure that absolutely does not want to be understood.
I am looking at a mess of registered agent history, address changes, public filings, old PDFs, and names that keep showing up in ways that may or may not matter. I paste a block of messy OCR text into Claude and ask it one specific question:
Map the timeline of address changes and flag any overlapping dates with these three individuals.
That is how I use AI most days.
Not to replace the work. Not to magically solve the investigation. Not to let a machine make judgment calls that I am responsible for making.
I use it to help me move faster through the mechanical parts of the work so I can spend more time on the parts that actually require judgment.
Every LinkedIn post right now seems to promise that AI will fundamentally change your life, double your income, replace your staff, fix your business, and probably fold your laundry if you prompt it correctly.
This post is not that.
This is just a look at how I actually used AI on a normal workday. No hype. No ten-step framework. No “future of work” sermon. Just a Tuesday in Kingsport.
The Morning: Deep in the Weeds
By 9:00 AM, Claude gives me a timeline of the corporate structure I am reviewing.
Some of it is genuinely impressive.
It catches a subtle address swap between two related entities that I probably would have missed on my first pass through the filings. Not forever, but on the first pass. That matters when you are staring at a screen full of stale addresses, registered agents, old business filings, and scanned documents that look like they were faxed through a toaster.
That one catch probably saves me 40 minutes.
But here is the reality check: it also confidently invents a middle initial for one of the directors because it merges two people with the same last name.
That is the part people do not talk about enough.
AI can be impressive and wrong in the same response. It can save you time and create a new problem if you trust it too much. It can find a pattern you missed and also make up a detail that has no business being in your report.
In investigative work, that is not a small issue.
If I had treated the AI output as finished intelligence instead of a working aid, I could have put bad information into a report. That is why my rule is simple: AI can organize, sort, compare, and structure. It does not get to verify itself.
I still do the reading manually.
I have tried the workflow where AI ingests every background document and summarizes everything for me. It sounds efficient. It looks good in a demo. It does not hold up for the kind of work I do.
The summaries are too clean.
They flatten the material. They strip out the strange details, the weird inconsistencies, the odd date overlaps, the one-off address, the name that appears once and then disappears. Those are usually the details that matter.
So now I use AI differently. I use it to turn raw information into tables, timelines, entity lists, and comparison charts. Then I go back to the source material and verify the pieces that matter.
That is less glamorous than “AI-powered intelligence,” but it is a lot closer to how the work actually gets done.
Mid-Morning: Consulting Without the Buzzword Fog
Around 11:00 AM, I switch gears.
Now I am working on an AI readiness assessment for a local business. Different kind of work, same general problem: too much noise, not enough clarity.
If you type “write an AI proposal for a manufacturer” into ChatGPT, you will get a polished pile of nothing.
It will talk about transformation, innovation, disruption, operational excellence, and probably synergy if you let it run long enough. It will sound like it was written for a conference booth in San Francisco, not a business owner in East Tennessee who is worried about uptime, payroll, customers, and whether this thing is going to create more problems than it solves.
That kind of AI output is exactly why a lot of people are skeptical of AI in the first place.
So I do not usually ask AI to write the proposal first. I ask it to attack the proposal.
Act like a skeptical plant manager in East Tennessee who has been running the same floor for 20 years. Read this project outline and tell me exactly why you think this is a waste of money and a threat to production uptime.
That prompt works.
It forces the model out of sales mode and into objection mode. It gives me the pushback I need before I put something in front of a real person who has every right to be skeptical.
On this particular day, the AI tears my draft apart. It points out that I had not clearly addressed downtime during implementation. It flags that the proposal talked too much about efficiency and not enough about risk containment. It calls out a section where I assumed the client had cleaner internal data than they probably do.
It was right.
I rebuilt the proposal around those objections.
But when it came time to write the actual executive summary, I did that myself.
That is another place where I draw a hard line. AI can help me think through structure. It can pressure test a plan. It can show me where I am being vague. But I do not trust it to understand the tone of a local business relationship.
There is a difference between technically correct and actually useful.
A lot of AI-generated business writing is technically fine and completely wrong for the room it is going into. Around here, people can smell fake polish. They do not want a TED Talk. They want to know what the thing does, what it costs, what it might break, and whether you are going to disappear after the invoice gets paid.
That part still has to come from me.
The Part I Keep Coming Back To
This is the core of how I think about AI in my own work:
I am not using AI to replace thinking. I am using it to handle the mechanical parts of thinking so I can spend more time on the parts that actually require judgment.
That sounds simple, but it is the difference between AI being useful and AI becoming dangerous.
Sorting a spreadsheet is mechanical. Extracting clause references from a long contract is mechanical. Turning messy notes into a clean outline is mechanical. Comparing two versions of a block of code is mechanical.
Deciding what matters is not mechanical.
Deciding what is true is not mechanical.
Deciding what should go into a client deliverable with your name on it is not mechanical.
That is the line I try not to cross.
Afternoon: Broken Code and False Confidence
At 1:30 PM, I am back in web developer mode.
A custom WordPress integration is throwing a fatal PHP error on line 42. It is one of those problems where I know I can fix it, but I also know it may take 30 minutes of staring at the same function before my brain catches the obvious thing.
I drop the snippet into ChatGPT and ask it to explain what the function is trying to do and why it is failing.
It spots the issue almost immediately: a deprecated hook and a bad assumption about the data being passed into the function.
Three minutes later, the problem is fixed.
That is one of the best use cases for AI in development. Not “build me an entire application while I drink coffee.” More like, “be a second set of eyes on this annoying function and explain what I am missing.”
Then, because this is usually how AI gets you, I get overconfident.
I ask it to write a custom regex to clean up some scraped data for another project.
It gives me confident garbage.
Not hesitant garbage. Not “you may want to test this” garbage. Confident, polished, absolutely wrong garbage.
If I had blindly pasted that into the workflow, it would have wiped out half the formatting I needed to preserve. Instead of saving time, I spent 20 minutes fixing the AI’s fix.
That is the rhythm of using AI for technical work right now. It can save you an hour, then cost you 20 minutes, five minutes later. The trick is not pretending it is always right. The trick is knowing where it is likely to be useful and where it is likely to sound smarter than it is.
Late Day: The Boring Wins Nobody Talks About
By 3:00 PM, I am doing the kind of work nobody writes thought-leadership posts about.
And honestly, this is where AI probably saves me the most energy.
I have a hideous CSV export with around 400 contacts. Names are inconsistent. Phone numbers are formatted six different ways. Some fields are combined. Some are blank. It is not hard work, but it is the kind of work that slowly drains the last functioning part of your brain.
I give it to an LLM and ask it to standardize the phone numbers, split full names into first and last name columns, and flag rows that need manual review.
Ten seconds.
That is not revolutionary. It is not going to be on a keynote slide. But it is useful.
Later, I have a long vendor agreement that I do not have the time or patience to read front to back at that moment. I do not ask AI to tell me whether the contract is good or bad. That would be a bad use of the tool.
I ask it for something narrow:
Extract only the clauses related to data retention, deletion, access rights, and confidentiality. Include the section numbers and quote the relevant language.
That is where AI shines.
Give it a narrow task. Make it show its work. Use it to reduce the pile of material you need to inspect, not to make the final call for you.
These are the boring wins. They are also the ones that matter most.
Cleaning up spreadsheets. Turning meeting notes into action items. Drafting a first version of a follow-up email. Finding specific clauses in a 45-page document. Explaining a PHP error. Reformatting ugly text. Building a table from messy source material.
None of that makes for a sexy AI demo.
But it gives you back pieces of your day.
What Actually Changed
The honest answer is not that AI made me a different person or completely changed how I work.
It did not.
I still read the source material. I still verify the facts. I still write the parts that require voice and judgment. I still have to know enough about the work to know when the AI is wrong.
What changed is that I now have a tool that can take some of the mechanical drag out of the day.
It helps me get to the real work faster.
That is not as exciting as the hype version. But it is more useful.
By 4:00 PM, I have used four different AI tools, closed two of them in frustration, saved probably two hours of work, and still had to manually verify anything that mattered.
That is most days.
No revolution. No magic. No robot coworker replacing human judgment.
Just a slightly more efficient Tuesday in Kingsport.