Implementing AI Solutions in Layman's Terms: A Recap of Our LA Tech Week Talk

The return on AI comes from the people in your business who build with it, not only from the people who use it, and that was the message John Gilligan, Chief Engineer at InfoNetworks, gave at LA Tech Week in Los Angeles on October 13, 2026.

His talk, "Implementing AI Solutions in Layman's Terms," was a working session for owners and managers, hosted by Ben Hakimi of InfoNetworks, a managed IT and technology services provider in Tarzana, California. The event page has the session's outline. This recap follows the talk in the order John gave it: an adoption strategy, a ladder of AI solutions, four things we built and where to start. He opened with a disclosure: AI was used to build the deck, just like everything else he showed.

Using AI is not the same as getting a return from it

The problem will sound familiar. A company buys everyone an AI license. People log in, some of them, some of the time. Nobody can point to money saved, or to anything new that got built. The tempting answer is to push harder on the people who are not using it. The talk argued that the effort belongs somewhere else.

The four levels of AI adoption

The talk sorts people into four levels. Levels 0 and 1 use AI. Levels 2 and 3 build with it, and that is where the return is.

Getting people from Level 0 to Level 1 raises the floor, and it is worth having. It is just not where leadership attention earns the most. Moving people from Level 2 to Level 3 raises the ceiling.

The reason is reach. A Level 1 who saves themselves some time every day makes one person faster. A Level 3 who builds a tool that saves the same time for everyone on a team makes the whole team faster. So the talk rates a builder by reach: how many other people their work lifts.

Five levers that move a Level 2 to Level 3

1. Tools: real builder tools and access to the data they need, not just a chat window.
2. Accreditation: a recognized title or certificate that says this person builds.
3. Visibility: put their work in front of the whole company. Demos, not thank-you emails.
4. Rewards: make building count in reviews, pay and promotion.
5. A Level 3 within reach: pair every Level 2 with a Level 3 they can call on.

Look past the IT department when you do this. Builders can be technical or not, and your best future builder may be in finance or operations.

A ladder of four kinds of AI solution

Next, the talk moved from people to solutions. Each rung up the ladder gives the AI more autonomy.

Efficiency AI and opportunity AI are different things

Efficiency AI does the things you already did, better: the same work and the same product in less time, such as drafting, summarizing and finding the answer faster. It is necessary, and everyone will have it soon.

Opportunity AI creates things that did not exist before: new products, new services and new ways to reach a customer. These are things that were not worth building until the cost of building collapsed, and this is where the advantage is. In the talk's words, efficiency AI sets the table, and opportunity AI is where we need to go.

Six checks for a good AI solution

• A real, recurring pain: something someone does every week and resents.
• Data you already have: the answer lives in a system you already own.
• A way to check the answer: if you cannot verify it, do not ship it.
• The lowest rung that solves it: do not build an agent when a workflow will do.
• Guardrails from day one: look before it touches, and a person approves anything risky.
• One owner: a named person, not a committee.

Connectors, skills and agents, in plain English

Three words came up throughout the talk.

• Connector (also called an "MCP server"): a secure doorway that lets an AI assistant look inside one of your business systems, with a written list of exactly what it is allowed to do. Think of a staff badge that only opens the rooms you need.
• Skill: a written recipe. Step-by-step instructions for a task you do often. Anyone can run it, and it does the task the same way every time, like the laminated checklist taped next to the machine.
• Agent: an AI assistant with a job. It follows the recipes and uses the doorways to finish a piece of work, then hands you the result to approve, like a capable assistant who knows where everything is kept.

Use, build, publish: what you need underneath

• Use: a chat assistant, such as Claude or ChatGPT in the browser or the app. That is enough for Level 1 and for Rung A.
• Build: a coding agent, such as Claude Code or Codex, running in a safe sandbox: a virtual desktop walled off from your live systems. Levels 2 and 3, and Rungs B to D, need it.
• Publish: a central, containerized app platform, run with your Level 3s or your IT team, that serves finished tools to everyone else. Anything other people will use needs it.

Chat alone will not take anyone from Level 1 to Level 3, because builders need a coding agent. Builders build in a sandbox, and your Level 3s and IT own the path to everyone else.

Four things we built, and where each sits on the ladder

Each of these was built by one Level 3. The clients stay anonymous here, as they did in the talk.

A real-estate client (Rung B: Applied)

In 24 hours we had a working demo: every building on one screen, with a chat box in the corner so the client could ask about their own data. In the demo, nobody looked at the charts. Everybody typed into the chat box. A chart tells you what happened; a good answer tells you what, why and what to do about it.

So we threw the dashboard away. The client already uses an AI assistant every day, so instead of asking them to open another app, we built a connector. Now their own property data answers questions such as "Who is behind on rent, and by how much?" or "Which repair requests have been open longest?" inside the assistant they already have open. Two lessons came out of it: a day to build means almost nothing to throw away, and it is better to bring the data to the people than the people to the data.

A healthcare call center (Rung B: Applied)

A healthcare organization running a patient call center was paying a lot per seat for software it was not happy with, and it had a wish list that included a live view of the call center, supervisors able to listen in and coach during a call, automatic scoring of call quality including tone, staffing forecasts against what actually happened, outbound calling campaigns and screen recording of agents' desktops. Our first reaction was "We cannot build that." We tried anyway, and had a working demo 24 hours later.

One engineer set direction and made the calls. One AI project manager planned the work, handed it out and checked it, and 15 AI coders worked in parallel, with automated tests written alongside the code. At one point two of the AI coders disagreed, in writing, about whose bug it was, and settled on a better fix without the engineer stepping in. The engineer's job was not typing. It was steering, and saying no.

Our own help desk (Rung C: Workflow)

In our own help desk, every ticket is researched before anyone opens it. When a customer emails or calls in a problem, the assistant reads the ticket and decides which specialist it needs: phones, email or network. It checks the customer's actual live systems through read-only connectors, finds past tickets that really resemble this one and what fixed them, and writes up what it found as a note on the ticket.

The human stays in charge. The assistant writes a note and a draft reply, and a technician reads it, decides and presses send. The assistant never talks to a customer on its own. We can also tell it what is going on today: if a phone carrier has an outage, we post one note to a board, and every ticket the assistant looks at after that takes it into account until the note expires. The return on this one was almost immediate, and it lifts every technician.

Security for one customer (Rung D: Agentic)

Every morning, for one customer, an AI security team starts work. An AI lead investigator plans the sweep, sends follow-ups and writes the report, while five AI investigators work in parallel on sign-ins, mailboxes, accounts and roles, files and sharing, and devices. The client's IT team gets one clear report a day, and a text message if something is urgent.

A person stays in charge where it matters. The morning sweep cannot change anything. To lock an account or reset a password, a staff member previews the change and types a code sent to their own phone. If a check fails, the report says it is incomplete; it never gives a false all-clear. Agentic does not mean unsupervised: it investigates alone, and a person approves the actions that matter.

The talk closed this section with the guardrails that make it safe enough to let AI into a business:

• Look, do not touch: almost every connector is read-only. It can tell you what is wrong; it cannot change anything.
• Passwords stay hidden: secrets are stripped out before the AI sees the answer. The one exception is restricted to named engineers and logged.
• Changes need a code: the few actions that do change something need a confirmation code sent to the person asking. No code, no change.
• Every change has a name on it: who asked, why and which ticket, written into the change itself.
• Each team sees only its tools: a help desk technician gets a reviewed set of tools, not everything.
• A human presses send: nothing goes to a customer without a person approving it.

Those guardrails are why our own engineers trust it, and why customers let us in.

Where to start: three things to do on Monday

Start with ideas. In the talk's model, anyone in any department, at any level, opens an idea skill and describes the idea in their own words. The skill asks what they do today and how often, where the information lives and what good would look like, then sorts the idea as a skill, a workflow, applied AI or agentic AI. If the idea is a skill or a simple workflow, the skill challenges the person to build it themselves, right now, and helps them do it. Bigger ideas go to a small review team of Level 2s and 3s, who decide which go forward and who builds them. Copies go to the person with the idea and to their manager, and the originator's name stays on the idea, whoever builds it. That is the accreditation, and the ideas show you who your builders are.

1. Give everyone a way to submit ideas. An idea skill anyone can use, with their name kept on the idea.
2. Find and grow your builders. The ideas show you who they are. Give your Level 2s a path to Level 3.
3. Pick one idea, and the lowest rung that solves it. A real recurring pain, data you already have and a way to check the answer.

Focus on your builders, build on the lowest rung that works, then climb.

John closed the way he opened: AI built the deck, and a human decided what went in it. The AI does the building. Your people decide what is worth building, who it is for and when to say no. Find those people and give them a ladder. As John put it at the end: go and find your Level 2s. If you would like a second opinion on where your first idea sits on the ladder, our consulting services cover strategic IT guidance and technology roadmaps.