A few months ago, I started thinking about the business problem differently.

The bottleneck at a solo consulting firm isn't skill. It's time. I'm one person doing development work, business development, marketing, client communication, proposals, and all the administrative overhead that comes with running a small business. Most of it has to happen in parallel.

The traditional solution is to hire people. Which makes sense at a certain scale. But for a solo firm that's intentionally staying lean, you're either outsourcing to freelancers — with all the coordination overhead that entails — or you're doing everything yourself and accepting that some things will get done poorly or not at all.

AI agents turned out to be a third option I hadn't fully considered.

What a "Custom Agent" Actually Is

When I say agent, I don't mean a chatbot you paste questions into. I mean a Claude instance with a specific system prompt that defines its role, gives it detailed context about the business, and sets the rules for how it should think and operate.

Think of it like a detailed job description for a very capable employee. Except instead of onboarding someone over three months, you write the system prompt once and it's immediately operational.

The key word is "specialized." A general-purpose AI assistant is useful. A specialized agent — one that knows the business, knows its domain, and has clear rules about what it should and shouldn't do — is a different category of useful.

The Agents Running at Reaction21

Here's what I actually built.

Marketing Expert

This agent writes LinkedIn posts, blog drafts, email campaigns, and content calendars. But it doesn't write generic content — it writes in my voice, following documented style rules I've built up over time. No corporate buzzwords. First person, always. Short sentences mixed with longer ones. Opinions stated directly. Dry humor when it fits.

It also knows the content themes that matter for Reaction21 — practical AI, .NET and Umbraco expertise, solo consulting lessons, project stories. It knows which phrases I'd never use ("leverage," "synergy," "digital transformation"). It knows the LinkedIn post formula I actually want.

The result isn't perfect on the first draft. But it's close enough that editing is faster than writing from scratch. And it never sounds like it came from a content mill.

CMS Specialist

This is actually the block builder skill I described in the previous post — but you could equally call it an agent. It knows Umbraco's block system cold, follows the project's conventions without being told, and handles the full workflow from element type creation to browser validation.

The distinction between "agent" and "skill" is somewhat fuzzy. Both are specialized AI context. The difference is mostly in how they're invoked.

Code Reviewer

GitHub Actions runs Claude automatically on every pull request. It reviews the diff, checks for accessibility issues, looks for patterns that violate project conventions, flags anything that looks like a potential bug. It leaves comments on the PR just like a human reviewer would.

It doesn't catch everything. But it catches a lot of the mechanical stuff — things like an aria-label that's missing, or a mapper that's using dynamic property access when strongly-typed access is required. That frees up my code review time for the things that actually require judgment.

Proposal Writer

When a new engagement comes in, the proposal writer has context on Reaction21's service catalog, pricing structure, approach to discovery, and what makes a good versus a bad fit. I feed it the discovery call notes and it produces a first draft.

Proposals used to take me 2-3 hours. The first draft now takes 20 minutes. The editing takes another 30. I've reclaimed something like 5-6 hours a month that used to go to proposal writing.

SEO/AEO Expert

This one focuses specifically on content optimization — not just for traditional search but for AI citation (AEO: Answer Engine Optimization). As AI search tools like Perplexity and SearchGPT become more common, the rules for how content gets surfaced are changing. This agent knows the current patterns for structured content, FAQ formatting, and the things that make content more likely to get cited by AI systems.

The Part That Makes It Actually Work

Individual agents are useful. But the more interesting capability is agents calling other agents.

When I ask the marketing agent to write a blog post about Umbraco, it knows it should pull context from the CMS knowledge base. When the proposal writer is generating an AI automation proposal, it knows it should reference the current service catalog and pricing.

This isn't theoretical. It's how complex tasks actually work in the real world — a specialist does their work, but they don't operate in a vacuum. They have access to context from other parts of the organization.

The coordination overhead that would normally come from getting three people to collaborate on one document? It goes away when all three are running in the same system.

What This Doesn't Replace

I want to be careful not to oversell this.

Agents are good at executing clearly-defined tasks within well-understood domains. The clearer the brief and the richer the context, the better the output.

They're not good at knowing what to work on. That's still me. I still set the priorities, make the strategic calls, decide what content themes matter this quarter, decide when a proposal should emphasize services differently. The agents execute. The judgment is mine.

They're also not good at anything that requires relationship context. A client call, a difficult conversation about scope, knowing that a particular client responds better to one approach than another — none of that gets delegated.

But for the structured, repeatable, domain-specific work that eats up a solo consultant's week? The difference has been real. Real enough that I'd call it one of the bigger productivity shifts I've had in the last five years.

For Other Small Businesses

If you're running a small professional services firm and you're wondering whether this applies to you — it depends on one thing.

How much of your week is structured, repeatable work that has clear inputs and outputs?

If the answer is "a lot" — proposals, reports, content, documentation, routine analysis — then specialized agents are worth exploring. The setup investment is real (writing good system prompts takes time), but it's a one-time cost that pays forward every week.

If your work is mostly relationship-driven, highly variable, or judgment-heavy? The ROI will be lower. That's just honest.

Interested in how AI agents could apply to your business? The first step is an AI Workflow Audit — mapping where your time actually goes and identifying where AI would genuinely help versus where it's just a distraction. Start at reaction21.com/contact.