The Honest Admission
I run a small team. For a long time, I felt like I had to apologize for that.
When a prospective client would ask how many people we had, I'd give a careful answer. Four people. Two full-time — myself and my wife, who manages operations — and two IT contractors we work with consistently. I'd watch their expression, waiting for the moment they decided to go with the bigger shop across town, the one with 40 employees and a glossy office in Irvine.
What I've come to understand is that the question "how many people do you have?" is really asking something else: "Can I trust that you'll be there when I need you? Will things fall through the cracks? Are you going to outgrow my problem?" Those are legitimate concerns. But headcount is a poor proxy for capability, and in 2024 and 2025, I rebuilt our operations in a way that makes me genuinely confident in how we answer those questions — not because we hired more people, but because we built something different.
We now operate with what I call an AI-native model. Alongside our four-person human team, we run 10 AI agents that handle specific, well-defined parts of our workflow every single day. This piece is my attempt to explain honestly what that means, why I built it this way, and what it means for the businesses that trust us with their IT.
What Changed: Integrating AI Agents Into Daily Operations
The shift happened gradually and then all at once. I had been using large language model tools informally for a while — drafting emails faster, researching a vendor we hadn't worked with before, sketching out a configuration approach before laying hands on the hardware. It was useful in the way that any good search tool is useful. But at some point in 2024, I started asking a different question: what if these capabilities weren't just a personal productivity shortcut but a structured part of how the entire business operates?
That question led to about six months of deliberate experimentation. I identified the parts of our workflow that were high-volume, repetitive, and documentation-heavy — the tasks that weren't hard to do but that were genuinely time-consuming, and where the cost of inconsistency was real. I then designed agents to handle each of those areas with defined inputs, defined outputs, and mandatory human review at the right checkpoints.
The result is a practice where our human team spends its time on the things that actually require human judgment — complex troubleshooting, client relationships, escalations, strategic decisions — while the AI layer handles the documentation, analysis, drafting, and research work that would otherwise quietly consume hours every week.
We call ourselves Southern California's AI-Native Technology Firm because this isn't a feature we bolted on. It's how we designed our operations from the ground up. You can read more about what that term actually means in our overview of the AI-native MSP model.
What the AI Agents Actually Do
I want to be specific here, because "we use AI" has become meaningless through overuse. Here is what our agents are doing day-to-day:
After a support session or project task, an agent generates structured documentation from notes, session logs, and resolution steps — saving to the appropriate client record in our system. What used to be intermittently completed is now consistently done.
We run RMM and monitoring tools across client environments. An agent processes alert data, identifies patterns across time windows, and produces a daily summary that flags what deserves a human engineer's attention and what is routine noise.
When a new CVE drops or a threat intelligence feed surfaces something relevant to our clients' stack, an agent drafts an initial analysis — what the vulnerability is, which clients may be affected, what the remediation path looks like. A human reviews before any action is taken.
For clients with compliance obligations — HIPAA, CMMC, FTC Safeguards — agents generate and maintain up-to-date compliance checklists, track control status, and flag items that are overdue for review. This keeps compliance a living process rather than an annual scramble.
Incident updates, maintenance notifications, and follow-up communications are drafted by an agent and reviewed by a human before sending. The result is faster turnaround and consistent professional tone — even at 10 PM on a Saturday when something breaks.
Our internal knowledge base — vendor-specific configuration guides, client environment notes, troubleshooting runbooks — is continuously updated by an agent that identifies gaps and drafts new articles based on recent support activity.
Monthly client reports covering uptime, patch status, incident summaries, and open items are generated by an agent using data pulled from our tools. A human reviews and personalizes before delivery.
When we're onboarding a new client or making significant changes to an existing environment, an agent reviews configuration documentation against our baseline standards and flags deviations for engineer review.
After a security incident or significant outage, an agent constructs a chronological timeline from logs and notes. This serves both the post-incident review and any client-facing documentation we need to provide.
When we're evaluating a new vendor, product, or technology for a client, an agent drafts an initial comparison covering key specifications, pricing tiers, and known limitations — giving the engineer a research head start rather than a blank page.
These agents run on large language model platforms, and they operate within structured workflows with defined outputs. They are not autonomous. Every output that touches a client — a report, a communication, a recommendation — goes through human review before it leaves our organization.
What AI Agents Don't Do
This section matters as much as the last one, so I want to be equally specific.
AI agents do not make client-facing decisions without human review. A draft is a draft. Every client communication, report, and recommendation is reviewed by a human before delivery. The agent accelerates the work; the human is accountable for it.
AI agents do not replace engineer judgment on complex problems. When a client's server is down at 2 AM, or when a firewall is misconfigured in a way that's blocking production traffic, or when something is behaving in a way none of the documentation predicts — that's a human problem. The engineer's pattern recognition, contextual knowledge of that specific environment, and ability to adapt to unexpected conditions are irreplaceable. The agent can surface relevant documentation; it cannot make the call.
AI agents do not provide accountability. This one is important. Accountability is a human attribute. When something goes wrong — and in IT, things go wrong — a client needs to be able to talk to a person who owns the outcome. That person is me. The AI layer handles workflow; the accountability layer is entirely human. Our clients have my cell number, not a ticket queue.
AI agents do not handle sensitive access or make configuration changes. The agents work with data and documents. They don't have privileged access to client systems, they don't push configuration changes, and they don't interact directly with production environments. Those actions require authenticated human access and deliberate human decision-making.
The honest version: AI agents are very good at high-volume, structured, documentation-heavy work. They are not good at ambiguity, novel situations, or anything that requires genuine judgment about what a specific client needs in a specific context. We designed our model around that reality — not the marketing version of it.
The Accountability Argument
Here's the claim that surprises people: because of how we've integrated AI into our workflows, we have more documentation and a more complete audit trail than most 10-person IT firms.
Think about how documentation typically works in a small IT shop. Engineers are busy. Tickets get closed without thorough notes. Config changes get made without a written record. Post-incident reviews are verbal, not documented. Compliance checklists exist as a PDF that nobody has opened since last year. This is not a character flaw — it's a resource constraint. Documentation competes with billable work, and in a small team, documentation usually loses.
Our AI-assisted documentation workflow inverts that. When a support session closes, the documentation happens. Not because someone had extra time, but because the agent generates it from what already exists — notes, session data, resolution steps — and puts it where it belongs. The engineer reviews it in 90 seconds rather than writing it from scratch in 10 minutes.
The result is that we have complete, timestamped records of what was done, when, why, and by whom. Client environments have living configuration documentation that actually reflects current state. Incident timelines are reconstructable from actual logs rather than memory. Compliance status is tracked continuously rather than assembled in a panic at audit time.
If a client ever needs to ask "what exactly happened on March 14th and why did you make that change?" — we can answer that question. That level of documented accountability is genuinely rare in a firm our size, and it exists precisely because we built the AI layer to make documentation a byproduct of the work rather than an additional task after it.
What This Means for Our Clients
Three things happen when you work with a team that operates this way:
Faster turnaround. When a monitoring alert fires at midnight, an agent has already produced the initial analysis by the time the engineer looks at it in the morning. When you need a report before a board meeting, it's not being assembled from scratch the night before. The prep work is continuous, which means the delivery is faster.
More consistent quality. One of the hardest things about a small team is that quality can vary depending on who's having a rough week, who's overloaded, or who happened to be assigned to a particular task. The AI layer creates a quality floor. Documentation follows the same structure every time. Reports cover the same items every month. Communications follow a consistent professional standard. Human variation happens above that floor, not below it.
Better documentation than you'd get elsewhere. I said this above, but it's worth repeating in client terms: your IT environment has a paper trail. If you ever want to move to a different provider, your documentation moves with you. If you have a regulatory audit, your records are current. If something goes wrong, we can reconstruct exactly what happened. That's a deliverable most clients don't realize they should be asking for.
The Competitive Position: Built-In vs. Bolted-On
The managed IT industry is going through an AI moment right now. Every RMM vendor is adding an AI feature. Every PSA platform is announcing a co-pilot. Every national MSP is issuing a press release about their AI transformation. I've read a lot of these announcements, and I notice a pattern: they describe AI as something being added to an existing operation.
We built our operation with the AI layer as a design assumption, not a retrofit. That's a meaningful difference — not because one approach is morally superior, but because the architecture is different. When AI is bolted onto an existing workflow, you get AI-assisted versions of your old process. When AI is part of how the workflow was designed from the start, the entire process is structured around what AI does well and what humans do well.
Our AI consulting practice actually grew out of this. Clients started asking how we operated, whether we could help them build something similar, and whether the same approach applied to their non-IT business processes. The answer to all three is yes — and it's a conversation worth having with our team if you're thinking about it for your own organization.
What "AI-Native" Doesn't Mean
I want to be clear about what we are not doing, because there's a lot of noise in this space and some of it is genuinely irresponsible.
"AI-native" does not mean we've automated away human responsibility. Every output the AI layer produces is reviewed by a human before it touches a client. We haven't reduced accountability — we've structured accountability more clearly, because the human review step is explicit and deliberate rather than happening (or not happening) somewhere in the middle of a busy afternoon.
"AI-native" does not mean we're cutting corners on cost and passing it off as innovation. Our pricing reflects the value we deliver, not the cost of avoiding human labor. The efficiency gains go into faster response times, more thorough documentation, and the ability to take on work that a 4-person team couldn't previously take on — not into underpricing the market and hoping nobody notices the difference in service quality.
"AI-native" does not mean AI makes the strategic decisions. When a client is choosing between on-premises and cloud infrastructure, or evaluating whether their security posture is adequate for their risk profile, or deciding whether a particular vendor relationship makes sense — those conversations are human conversations. I have them, my contractors have them, and the AI layer has nothing to do with them.
And "AI-native" does not mean we're using AI as a crutch to avoid developing real expertise. The agents are useful precisely because we understand the domains they're working in well enough to know when their output is right, when it's almost right, and when it needs to go back to the drawing board. An agent that generates a compliance checklist is only as valuable as the human who knows enough about that compliance framework to catch an error in it.
The Challenges and Limitations — An Honest Reflection
I would be doing you a disservice if I wrote this as though everything works perfectly all the time. It doesn't.
The most significant ongoing challenge is output quality management. AI agents produce drafts, and drafts require review. When the team is busy — really busy — the temptation is to spend less time on that review step. We've had to build explicit checkpoints into our workflows to prevent that from happening, because the moment you start rubber-stamping AI output is the moment the quality floor becomes the quality ceiling.
There is also the challenge of keeping agent context current. Agents work from the information they're given. If a client's environment changes in a way that doesn't get documented — a new server added, a software update applied outside our workflow, a configuration change made by someone else — the agent is working from a stale picture. We've addressed this through tighter change management protocols, but it requires discipline to maintain.
The human relationship layer requires more intentional effort, not less. When agents are handling documentation and drafting and reporting, it can feel like client communication is "covered." It isn't. Clients need to hear from a human who knows their specific situation and cares about their outcome. We schedule proactive human touchpoints specifically because we know the agent layer doesn't replace that.
And finally: this model requires continuous refinement. The agents we run today are better than the ones we ran six months ago, and they'll be better again in six months. That's good, but it means the work of building and maintaining this system is ongoing. It's not a one-time implementation. For a small team, that ongoing commitment is real work — valuable work, but work.
Work With a Team That's Transparent About How They Operate
I wrote this because I think prospective clients deserve to know how we actually work — not a sanitized version of it, not a marketing story, but the real operating model with its real capabilities and real limitations.
The businesses that are best suited to work with us are ones that value consistency, documentation, and transparency. If you want to understand exactly how your IT is being managed, exactly what happened during a particular incident, and exactly what your compliance posture looks like at any given moment — we built our operation to give you that. If you'd prefer a traditional relationship where you don't think about IT until something breaks, we're probably not the right fit, and I respect that.
We're based in Corona, California, and we serve businesses across Southern California. Our team is small by design, not by default, and the combination of human expertise and AI-augmented workflows is how we deliver a quality of service that our size alone wouldn't support.
If any of this resonates — or if you have questions about how it works in practice — I'd genuinely welcome the conversation.
Ready to Work With a Team That's Transparent About How They Operate?
We're happy to walk you through our operating model, show you what documentation looks like at the end of the first month, and have an honest conversation about whether we're the right fit for your organization.
Start the Conversation