How we work
The engagement from the first conversation to handoff, and what you get at each step.
What we believe
Better work is the point. Your people’s attention is what it buys back.
Richardson Applied AI builds working AI systems around how an organization operates. These are the beliefs I built the practice on and the rules we work by, and they decide what we build, what I’ll tell you not to build, and how we talk about results.
Why
Six beliefs behind the work
I came to these by building systems inside real operations and by running my own company on the same method.
Judge AI by the work it produces.
When these tools are set up around how your organization works day to day, the work itself gets better: a quote that comes out complete with the reasoning written down, a report that’s more thorough and consistent, an answer that’s better researched and easier to defend. Speed often comes along with it, and it’s the smaller part. I judge the technology by the quality of the result and the effort it takes to get there, review and corrections included.
The leverage compounds.
In the places where it fits, the return is out of proportion to what it costs to get there, and it doesn’t stay flat. Every process your people write down and check makes the next build easier, because the next one starts from knowledge you already captured. So the first project is also the start of the record every later project builds on.
Context is the job.
A capable model still needs the right context, and a lot of that context lives with the people who do the work: the reasons behind a price, the customer who always revises the print, the rule nobody wrote down because everyone knows it. The model is the easy part of this. Getting that knowledge into a form the software can use is most of the work, and the people who know the job stay involved in checking that the system understood it.
The edge doesn’t go to whoever funds the frontier, and it doesn’t go to whoever owns the model. It goes to whoever connects the technology to a real problem set.
Humanity is the point of the leverage.
When the document work is handled well, what comes back is your best people’s attention: the judgment call on a hard quote, the customer on the phone, the conversation across a desk that decides whether a deal closes, a donor renews or an employee stays. That attention is the first thing to go when everyone is buried, and giving it back is the reason I do any of this.
You own what we build.
The systems and the knowledge behind them live in your accounts, under your logins, in plain files your company owns. Your passwords stay in your password manager, and at least two of your own people can get into everything without us. The engagement is designed to end with your people running it. Cadence, our ongoing support after the handoff, is optional and month to month, so you can cancel any month and everything keeps working.
Swap the AI; the Standard stays.
Model prices and products change month to month, and in August 2026 the API market repriced three times in four weeks (State of AI, September 2026). A system that only runs on one vendor’s platform is a liability sold to you as a feature, so your Operating Standard is written in plain language any capable model can follow, and the link to any one vendor is a thin layer on top. When a better model shows up, we test the switch end to end before it counts, and you keep everything you built.
The Standard is the fixed part, kept as plain files you own. Model A, B or C plugs into the same interface, so changing models doesn’t mean starting over.
How
The rules we work by
These are written into the method we use on client work and on our own company, so you can hold us to them.
- 01
The model reads and drafts. Code decides. People approve.
The AI does the reading and the first draft, where messy input needs interpretation. Exact calculations, comparisons and rules go in ordinary code that can be tested, and a person approves anything that leaves the building.
- 02
Gate what can’t be undone, and stop gating everything else.
Money leaving the building, a message in front of a customer and a record you can’t restore all wait for a person, permanently. Drafts and staged changes that can be thrown away don’t need a signature on every step, and review points come off only as the system proves itself and the owner signs off in writing.
- 03
Count the net honestly.
A system is worth what it produces minus what it costs to keep: maintenance, review, corrections and people’s attention, as well as the model bill. If the upkeep eats the gain, it’s an experiment, and we’ll report it as one.
- 04
Start with one real process.
Pick a process your team understands and would notice getting better, agree on what a better result looks like, and measure how it runs today. No result gets claimed without that baseline.
- 05
Prove it small, and stop if it can’t work.
When we aren’t sure something will work, we test it first on your real documents in a small, paid step that ends with a plain decision to go, change course, or stop. You don’t fund a big build on a guess.
- 06
Build the smallest thing that will last.
If the system you already use can do the job, the work stays there. A written procedure, a scheduled job or a small app comes in only when it’s needed, and we don’t build anything your people can’t keep running.
- 07
Done needs evidence.
Done means it ran on its own schedule, someone checked the result where it landed, and your people accepted it. A good demo doesn’t count, and neither does an AI reporting that the job went fine.
- 08
Reality wins, and the record is fixed the same day.
Your Operating Standard holds how the work is done, with every line sourced and dated. When the record and the work disagree, the work wins and the record is corrected that day, and keeping it current is part of doing the work.
- 09
Every number carries a source.
We don’t invent savings or dress up case studies, and a client’s results stay private until the client has checked them and agreed to share them. That’s why you won’t find client numbers on this site.
- 10
You’ll get a straight answer, including no.
Sometimes the most useful thing we can tell you is that AI won’t move anything for you yet, and what would change that. Every engagement is scoped independently and quoted directly. The first conversation costs nothing, and when there’s a real path, the working session after it is typically free too.
In practice
Where these show up in the work
The Operating Standard
What we install: a sourced, dated record of how your work gets done, and the people who keep it current.
State of AI
Our monthly read on where AI stands for people who run companies, with every number linked to its source.
Conversations are always free.
Tell us which process keeps stalling, and we’ll tell you whether AI can move it, including when it can’t.