AI Agents vs. Hiring: What Pays Off for a 20-Person Firm

Picture a 20-person consulting firm that spent three months debating whether to hire a dedicated operations coordinator or try something else. Their back office was creaking. Proposals were going out late, client follow-ups were falling through the cracks, and the two senior partners were spending roughly a day each week on work that felt, in their own words, administrative. They eventually chose to deploy a small stack of AI agents instead of hiring. Six months later, the picture looked meaningfully different, though not in every way they expected. Here is what that decision actually involved, and what firms of a similar size can learn from it.

The Problem That Prompted the Decision

The firm handled strategy engagements for mid-market manufacturers. Their revenue was healthy but their processes were not keeping pace with growth. Proposals required pulling data from four different sources, formatting it into a branded template, and routing it through two approvals before it could go out. Client onboarding involved a checklist of about thirty steps, most of which lived in someone's memory. And weekly reporting to clients meant one analyst spending most of Friday pulling numbers and writing summaries.

None of these tasks required deep judgment. They required consistency, attention to detail, and time. That is exactly the profile where AI agents tend to perform well, and it is also exactly the profile that makes hiring feel expensive relative to the output.

What AI Agents Actually Did in This Context

The firm worked with an implementation partner to deploy three agents handling distinct workflows. The first handled proposal assembly, pulling scoped data from their CRM and project management tools, populating a template, and flagging the draft for human review before it went anywhere. The second managed client onboarding communications, sending sequenced messages, tracking responses, and escalating to a human if a client went quiet for more than forty-eight hours. The third generated the weekly client reports, pulling live data from their dashboards and drafting a narrative summary that an analyst would then review and personalise in about fifteen minutes rather than three hours.

None of these agents made final decisions. They handled the mechanical construction of work and surfaced it for human sign-off. That distinction matters more than most articles acknowledge. The value was not in replacing judgment. It was in removing the prep work that preceded judgment.

Where Hiring Would Have Been the Better Answer

It would be dishonest to frame this as a clean win for AI across the board. There were two areas where the firm eventually did hire, and where an agent simply could not replicate what was needed.

The first was client relationship depth. One of the partners had been handling all senior client calls alone, and as the firm grew, this became a bottleneck that no amount of automation could fix. They needed a senior account director who could build trust with clients, read a room, and exercise genuine discretion. That hire happened four months in and was absolutely the right call.

The second was internal culture and coordination. At twenty people, teams still need a human who owns the messy, relational side of keeping things running, particularly during growth. An operations coordinator role that might have been purely administrative a few years ago has evolved. The firm found that what they actually needed from that role was judgment-heavy enough that a human was clearly the right answer. What the agents removed was the administrative load that would have consumed most of that person's time, making the role more strategic and easier to hire for.

The Real Cost Comparison at This Scale

At twenty people, budget scrutiny is intense. A mid-level operations hire in most markets will cost somewhere in the range that makes small firms genuinely pause. AI agent infrastructure, by contrast, tends to sit at a fraction of that annually, though costs vary considerably depending on the tools, the integrations required, and whether you use an implementation partner or attempt to build in-house.

What the firm found more useful than a direct cost comparison was thinking about time recovery. When they mapped out how many senior-level hours per week were being consumed by work that required no senior judgment, the number was uncomfortable. Recovering those hours and redirecting them toward billable work or business development had a value that was straightforward to calculate. That framing helped them justify the investment internally without needing to pretend they had perfect cost data.

What the Decision Process Actually Looked Like

The firm did not arrive at this approach through a clean strategic process. They started with a frustration, looked at their options, and made a provisional decision to pilot agents on one workflow before expanding. That pilot mindset was probably the most important factor in making it work. They did not commit to a full transformation before they knew what they were dealing with.

They also did not treat the AI agents as a set-and-forget solution. The proposal assembly agent required meaningful refinement over the first two months as edge cases emerged. Someone had to own that process, and in practice it was a senior analyst who developed an interest in the tooling. Firms that deploy agents and assume they will operate cleanly without ongoing attention tend to be disappointed.

What Firms Like This Should Actually Weigh

If you are running a firm of roughly this size and facing a similar decision, the most useful question is not whether AI agents are impressive. They often are. The question is whether the workflows you are trying to fix are actually agent-appropriate. That means they are repeatable, they have clear inputs and outputs, the stakes of an error are manageable, and a human reviewing the output can catch problems before they reach a client.

If your bottleneck is high-judgment work, relationship management, or anything where context shifts unpredictably, hiring is almost certainly the right answer and agents will not save you. If your bottleneck is volume, repetition, and the mechanical assembly of information that already exists in your systems, agents can return real time and real money to your team.

The firm in this story did not choose between AI and people. They used agents to change what their people spent time on, and then made better hiring decisions as a result. That framing, more than any headline comparison, is probably where the real value sits for firms at this scale.