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

When a 20-person architecture firm in Bristol started losing bids because proposals were taking too long to produce, the managing partner faced a familiar crossroads: hire another project coordinator, or try something different. What followed over the next six months became a practical education in where AI agents genuinely earn their keep and where they fall flat on their face.

The Problem That Started the Conversation

The firm was producing roughly 40 client proposals a year. Each one pulled a senior architect away from billable work for two to three days. Multiply that out and you have a significant chunk of revenue-generating time consumed by document assembly, formatting, and chasing down boilerplate copy. The instinct was to hire a dedicated bid coordinator at around £35,000 to £40,000 a year. Before signing a contract, though, the managing partner agreed to a three-month test with an AI agent workflow instead.

The agent was configured to pull from existing project case studies, adjust tone based on client type, and draft a first-pass proposal document that a senior architect could then review and personalise. It did not replace human judgment on strategy or relationship nuance. It handled the scaffolding.

What the Numbers Actually Looked Like

The AI agent setup, including a subscription to the underlying platform and about two days of configuration work, came to just under £4,000 upfront. Ongoing costs sat at roughly £300 a month. In the first quarter of use, the time spent by senior staff on first-draft proposals dropped by around 60 percent according to their own tracked hours. That freed capacity translated into one additional billable project being taken on that quarter.

A full-time hire would have cost the firm considerably more once you factor in employer national insurance, pension contributions, equipment, onboarding time, and the unavoidable ramp-up period before a new employee is genuinely productive. That is not an argument that AI agents always win the cost comparison. It is an argument that the comparison needs to be honest and fully loaded on both sides.

Where the AI Agent Actually Struggled

The firm hit real limitations quickly in two areas. The first was client calls. Prospective clients who rang with nuanced briefs needed a human who could read the room, respond to hesitation, and build trust in real time. No agent handled that and none was expected to. The second limitation was internal coordination. When a project hit a complication mid-delivery, the team needed someone who could physically walk across the office, read the mood, and make a judgment call. An agent working on defined inputs and outputs could not substitute for that kind of contextual human presence.

There was also an early quality problem. The first few proposals the agent produced were competent but generic. It took deliberate effort to feed the system enough firm-specific language, past project details, and client personas before outputs started sounding like the firm rather than a template. That investment of time is real and small businesses often underestimate it.

The Honest Case for Hiring Instead

For this firm, there were tasks where a hire would have been the right answer. If the primary gap had been relationship management, business development calls, or on-site client liaison, no AI agent in its current state would have covered it. There is a category of work that is fundamentally about sustained human presence and accountability, and firms of 20 people often need more of that, not more automation.

The AI agent made sense specifically because the bottleneck was a well-defined, repeatable task with clear inputs: project details go in, structured proposal comes out. If your bottleneck is something messier, like strategic account management or creative direction that requires genuine back-and-forth with clients, then hiring a skilled person is not just preferable, it is the only sensible option. AI agents work well on processes. They are poor substitutes for roles that are primarily about judgment and relationship.

How a 20-Person Firm Should Frame This Decision

The most useful question is not "can an AI agent do this job" but "what is the actual shape of the gap we are trying to fill." If the gap is a repeatable process that currently eats senior staff time, an AI agent workflow deserves a serious look. If the gap is a human role that requires presence, accountability to clients, and ongoing professional judgment, then hire.

For firms at the 20-person scale, cash flow also matters in a way it does not for larger businesses. A bad hire at this size is genuinely disruptive. A badly configured AI agent is cheaper to unwind. That asymmetry of risk is worth considering even if the cost comparison looks close on paper. On the other hand, a great hire grows with the firm and brings initiative that no current AI system replicates. The firm in Bristol eventually did hire a bid coordinator twelve months later, but by that point the AI agent had already cleared enough time to make that hire comfortable rather than urgent.

What This Means for Your Firm Right Now

If you are running a firm around this size and wrestling with the same question, the practical starting point is to map out where senior time is actually going each week. Not where you think it goes, but where it actually goes. Often the tasks most ripe for AI agent support are not the ones people assume. Inbox triage, meeting summaries, first-draft content, data extraction from documents, and internal reporting are all areas where agent workflows have shown consistent value in small professional services businesses.

The mistake to avoid is treating AI agents as a cost-cutting move designed to avoid hiring people. That framing leads to bad decisions in both directions. The better frame is capacity. Where is your existing team bottlenecked on work that does not require their specific expertise, and could an agent clear that path so the people you already have can focus on the things only they can do. When firms approach it that way, the question of hiring versus AI agents becomes less of a competition and more of a sequencing problem.