YOUR BUSINESS AUTHORITY

Springfield, MO

Log in Subscribe

PAID ADVERTISING

Meet Jonathan Hodges, AI Software Development and Custom Solutions Expert

Posted online

Jonathan Hodges is the Vice President of Technology at Classy Llama, bringing over 15 years of experience in web development, software engineering, and leadership within the ecommerce and technology sectors. With a strong foundation in mathematics and computer science from Evangel University, Jonathan has a proven track record of forming and leading teams, implementing Agile methodologies, and optimizing business operations to deliver impactful results.

Jonathan has held several key roles at Classy Llama, including Director of Development and Chief Operating Officer, where he demonstrated expertise in technical services, key performance indicators, and collaborative problem-solving. His skill set spans web applications, PHP, JavaScript, AI tooling, and product strategy, bolstered by certifications like the Magento Developer Certification Plus and Databricks Lakehouse Fundamentals.

A dedicated leader and innovative thinker, Jonathan is passionate about leveraging emerging technology to create value-driven solutions, improve operational efficiency, and foster community engagement in the tech industry.


Can AI answer questions from everything my company knows?

Yes. People call this a "company brain," and it comes down to something AI is already very good at: taking in a large amount of information quickly and putting together an answer from what's available. As it gets better at that, it earns a bigger place inside a company.

There are a few steps to getting there.

1. Map what your company knows Start by mapping all the data in your organization. Where does everything live? How do messages come in? Where are the SOPs? What is the canonical version of your company, the one everyone agrees is right? That can include things like:

• How we build out a proposal.
• How we run a sales call.

• How we send out a marketing message.

• The client journey from beginning to end.

• The onboarding journey from beginning to end.


Anything the AI would need to find should be mapped, and written down where someone can find it.

2. Connect those sources and tell the AI what takes priority
Next you hook those sources up so the AI can read them. That can be as simple as "here are all the documents in our company that are public or open to everyone inside it." It can also get as involved as access control for individual users and groups, with a process set up for each department. A company brain can be as simple or as complicated as your company is, and you don't have to start with the whole thing.

What does matter is starting with real data, and telling the AI where that data is and which sources win when two of them disagree. If one document says the old price and another says the new one, it needs to know which one to trust.
If you do get to permissions, put them in the system underneath the AI, where each person's access is checked before anything is searched. Instructions to the AI like "don't show payroll to the sales team" are much easier to talk your way around.

What it looks like when it works

Done well, you get very fast answers to questions and better decisions, because each decision can take in large parts of the company, based on information your company actually has. Done poorly, you get mediocre results. In my experience the difference is a good map, and access to the information both today and as it changes.

Where to start

The question I ask myself is: what part of the company can I map easily and quickly that will help the most people? I don't start with the hardest problem in the company. I start with the one that's easiest for AI to solve. A better candidate is something a computer finds easy and people find hard, like looking across a lot of well-documented areas to find an answer. A person can't hold all of that in their head, and the AI can read all of it in a few minutes. The mapping is the part you can start on your own this week. Getting the connections, the permissions and the testing in place takes more than talking to an AI, so plan on having someone build it with you and review it before your team starts relying on it.


Can one person really run a team of AI agents?

Yes, I do it every day. The agents generally keep up fine. Where I ran into trouble was keeping up with reviewing everything they produce.

Over the last month I went from working with two or three agents at a time to working with five or more directly and another five to ten in the background. Most of my work runs through Claude Code, Anthropic's agent tool, on my own machine. I was used to having about five sessions open on different screens and switching between them, and I knew exactly where everything was. Then I found out Claude Code could run agents in the background and realized I could get a lot more done. The problem was I couldn't find anything anymore, so I started building a system on top of it, which I call Fleet, to keep track of them all.

Here's basically how it's set up today:

1. Each client project has a manager agent. I tell it what I want, it writes up a work order, and it hands that to a builder agent that does the actual work and reports back.

2. A fleet manager agent watches every session and ranks what needs me right now, so I'm not hunting through windows trying to find the one that's stuck.

3. A planner agent reads my email and calendar, puts together my week, and asks me questions when it can't figure something out on its own.

4. Every piece of work is a task with a due date. The system only closes a task on its own when it can point to evidence the work is actually done.

What broke

Adding the manager agents pretty much blew up my workflow. Everything broke for about two weeks while I figured out how to keep up with agents that were starting other agents without me watching. At one point I had about 300 open items in front of me. The agents were doing their jobs fine and I was the bottleneck.

The worst part was rereading. An agent would hand me a 20-page document, I'd ask for a small change, and it would give me the whole thing back. By the fifth time through I was reading 20 pages to find two sentences that changed, and I was probably spending about as much time as if I'd written it from scratch. The quality was better but it was exhausting.

I think most businesses are going to run into this over the next year. AI work keeps getting cheaper and the time it takes you to check it stays about the same. Every agent you add is another stream of work somebody has to read, approve, or fix, and that's usually the owner or a manager who already has a full day.

What fixed it

Most of the fixes came down to protecting my attention. I've come to realize it's the most important thing I have and I can't make more of it. These four changes did most of the work:

1. When a draft changes, I only see the change. Every version is kept, and when a new one comes in it shows me what's different from the last one I read. That changed the way I work with AI more than anything else I built.

2. I read at the level of detail the decision needs. Each agent's work gets summarized a few ways: a couple of sentences, an outline, more detail under each point, and then the full transcript if I want it. I spend most of my day in the outline, and when my eyes get tired it'll read the outline aloud to me.

3. The writing comes out in my voice. Drafts go through a pass that rewrites them using real examples of my own writing, because describing your style to the AI doesn't work very well. A study published this year found that when a model was given a written description of an author's style, its imitation passed a style check less than 7% of the time. With two real paragraphs from the author it passed 91 to 100% of the time.

4. Tasks get closed with proof. The system checks its open tasks against what was actually done and closes the ones it has evidence for. That took me from about 300 open items to about half that.

I also have a rule for myself that's more about the quality I require than about handling more agents. Nothing goes out without me reading it. Every email, proposal, or client document an agent writes lands in one place as a draft. I read it, accept a version, and send it myself. No agent sends anything on its own.

It doesn't all feel easy yet. It's a lot better than it was, and I'm finally at the point where it's saving me real time.

If you're not building your own

You probably don't need to build what I built. The same problem shows up the first time anyone on your team starts using AI agents seriously. Here's what I'd do:

1. Budget the review time before you add the agent. If an agent produces ten reports a week, somebody has to read ten reports a week. Figure out who that is and how long it takes them.

2. Decide what needs a person's approval. For most businesses that's anything a customer, vendor, or regulator will see. Write that down before you turn anything on.

3. Ask for a summary first and the full document second. Most AI tools will give you a short version if you ask. Read that, and only open the long version when the decision is worth it.

4. Keep every version. When someone asks the AI for changes, keep the before and after so the reviewer can see what moved. A shared document with version history works fine for this.

5. Give it your real writing. If you want AI to sound like you, paste in three or four emails you actually sent instead of a list of adjectives.

6. Be patient with the slower, more powerful tools. The AI built into a lot of business software is tuned to answer in a few seconds, partly because that's what a cheap per-seat price can pay for. In my experience, waiting ten minutes for a finished answer usually beats waiting 30 seconds for something you then have to ask ten more questions about.

Where this is headed

This goes well beyond my own desk. A sales team that wants to ask questions across its CRM, email, and call recordings needs agents that can reach those systems, permissions so each person only sees what they're allowed to, and a way to review the answers that doesn't eat up the afternoon. That's a big part of what I'm building for businesses right now.

If you're just starting out, start with one agent and one job. Pay attention to how long it takes you to check its work, because that tells you how many more you can take on.


Should I buy an off-the-shelf AI tool or build something custom?

I've spent most of my career building custom software, so you'd probably expect me to say build. My default is still buy, and buy almost every time.

The test I use is pretty simple. If the job you want AI to do is something every business does, like answering email, scheduling, or chasing invoices, somebody has already built a tool for it and priced it at a fraction of what custom would cost. Buy it, and be willing to bend your process a little to fit the tool. In my experience most of the "we do it differently" in those areas is habit, and it isn't worth paying to preserve.

Build when the thing you want automated is the reason customers pick you over the shop down the road. Maybe it's how you quote a job, or the way you decide which leads get a call back. Nobody sells that off the shelf, because nobody else does it your way, and if you try to buy it you'll spend months fighting a tool that was designed around somebody else's business.

The cost side has changed. With AI writing most of the code, a focused build that would have cost six figures a few years ago now lands closer to half or better. That lowers the bar for building. What hasn't changed is that you have to keep custom software alive once you own it: hosting, updates, security patches, and somebody who understands it when it breaks. Budget for that before you build.

A practical way to settle it: try to buy first. Spend two weeks with a couple of off-the-shelf tools and see where they rub. If you're fighting the tool over the part of the work customers actually pay you for, that's your build signal. Often that ends up as a hybrid: keep the tool for the generic parts and build only the one piece it can't do. If you're fighting it over something generic, change the process instead. Most businesses I talk to end up buying, and that's generally the right answer.


How can AI help with my email?

Every morning AI goes through my inbox before I do. It reads each thread all the way through, files the receipts and newsletters, and for the repetitive stuff (scheduling, pricing questions, the follow-up I've answered some version of twenty times) it leaves a draft reply sitting in my drafts folder, written pretty close to how I'd write it myself.

Then I do my part. I read each draft, fix what's off, and hit send. Nothing goes out without me looking at it. Every so often a draft misses, maybe it offers a meeting time I don't actually have open, and catching that is still my job. Most mornings the whole inbox takes ten minutes of my time instead of the hour it used to.

You don't need my setup to get most of this. Take the five emails you answer the same way every week, paste a couple of your real past replies into a tool like Claude or ChatGPT, and ask it to draft the next one. Give it actual sent messages, since it picks up your voice from examples far better than from you describing it. In my experience the drafts get usable within 1-2 weeks of correcting them. Two cautions. Keep sensitive information out of consumer AI tools: customer financials, health info, and anything else you wouldn't want leaving the building. Business-grade accounts handle privacy a lot better, and it's worth an hour to get those settings right. And don't let it send on its own. The drafting is where the time savings are anyway, and I'd rather spend thirty seconds reviewing than have to explain an email I never read. Email turns out to be a pretty good first test of AI in general, since you already know what a good reply looks like and can check its work in seconds. Most people I've walked through this see the payoff in the first week.

I keep hearing AI can now "work on its own." What does that mean for my business?

Recently I sat down with a folder of recorded sales calls from a local business here in southwest Missouri. The owner wanted to know why some customers stay and others leave. The old way to answer that is somebody listening to hours of calls, taking notes, and burning a week they didn't have. Instead, I handed the folder to AI and told it what I was after. It transcribed every call, went looking for patterns, hit a few dead ends, backed up and tried different angles, and checked its conclusions against the calls before showing me anything.

When I came back, a finished analysis was waiting. The customers who stayed weren't the ones whose rep followed the script best. They were the ones whose rep sounded like they actually cared. That's something the owner can coach to and hire for. This kind of AI is called an agent.

Instead of answering one question and stopping, an agent keeps working toward a goal: it makes a plan, uses tools, tests its own progress, and gets itself unstuck along the way. If you've tried AI and gotten nothing like this, there's a reason. Two things drive the quality of the result, and most people only know about one of them.

The first is the model, the AI brain itself. The second is the harness, the tool wrapped around that brain that lets it read your files, run software, and check its own work. The same question asked in a free chatbot and in a good harness will give you night-and-day different results.

For the analysis above, my setup was Anthropic's Claude Opus model running inside Claude Code, a developer tool. That combination is the best I've used. If you're not technical, Claude Cowork gives you the same kind of power without the setup: point it at a folder of documents on your computer, give it instructions and permission to work, and check back when it's done. The models keep improving too. Anthropic's newest,Fable, runs longer, gets itself unstuck better, and keeps testing its work as it goes.

OpenAI and Google both have capable models and agent tools as well, and they're worth trying if you're already in those ecosystems. Claude with a good harness is simply what works best for me right now. My advice: give AI one real project this month. A stack of customer feedback, a messy spreadsheet, a pile of old proposals. Spell out what "done" looks like. Then review the work like you'd review a new hire's first project, because it can still take a confident wrong turn, and catching that is still your job. Most business owners I meet aren't overusing AI. They're asking it for far too little.


How do I keep up with the pace of change in AI?

AI is moving fast, and it is easy to feel behind. The key is combining broad awareness with a few focused deep dives and ongoing usage.

Use newsletters like The Rundown AI, TLDR AI, and Bay Area Times. Spend about 10 minutes a day scanning for what is worth exploring.

Then go deeper 1 to 4 times a week with videos, articles, or blog posts that matter to you.

Finally, get hands-on. Try tools as you learn, follow along, and stay curious. This often pays for itself in time saved.

The key is consistency. Keep learning and stay intentional.


What’s the ROI of investing in AI?

While ROI can vary, I’ve seen two types of initiatives consistently deliver strong returns. First, focused upskilling of interested or high-performing team members. If someone is eager to use AI, teach them how and give them access to the tools they want. You’ll see productivity gains that far outweigh the cost. Second, targeted programs. Look for repetitive or time-consuming work that AI can handle well. Bring in an experienced AI implementer with a clear ROI focus. These programs often pay for themselves within a year or less.

 

What is the most significant recent AI release I should know about?

Most people have used ChatGPT or Gemini, but a growing number are discovering Anthropic's Claude, especially for on-computer work. Claude Code took the programming world by storm, and to extend that beyond coding, Anthropic built Claude Cowork. It reads and writes from a folder on your computer, handling Excel sheets, PDFs, Word documents, and more. Drop in your documents, give it instructions, and walk away. It's the force multiplier people have been hoping for. If you haven't tried it, sign up for Claude and give it a spin.


What skills do I or my team need to work with AI?

To get real value from AI, curiosity matters most. Experimenting unlocks learning and ideas, creating a cycle of discovery.

Working with AI is like working with a team. Simple questions or shallow delegation produce limited results. Clear communication, strong leadership, and strategic thinking lead to better outcomes.

Instead of asking, “How can I make a lot of money?” try, “What are five businesses I could start that would succeed today? Use research, what you know about me, and interview me to understand my skills.”


Can AI integrate with my existing (CRM, ERP, etc.) systems?

Yes! AI is most powerful when it’s deeply integrated into your company’s workflows. Many modern CRMs and ERPs now include native AI capabilities, though their quality varies widely.

I work directly with companies to integrate AI into existing systems, automating workflows in part or end-to-end to reduce slow, manual processes. The most successful integrations start with human oversight and evolve as confidence grows, gradually reducing manual involvement to only the most complex cases.

Comments

No comments on this story |
Please log in to add your comment
Editors' Pick
Open For Business: 9 Lives Tavern

Shortly after getting married, Amos and Shelby Hale became first-time business owners with the July 10 launch of 9 Lives Tavern in downtown Ozark.

Most Read
Update cookies preferences