Scope note: This page covers custom AI agent builds — the highest-complexity implementation tier. For simpler workflow automation, see C18 AI Automation Build. For knowledge systems and custom GPTs, see C19 Custom GPTs. If you are unsure which tier is right, start with the C16 AI Tools Assessment.
"The decision between automation and an agent is simpler than it sounds. If you can fully map the process before it runs — every step, every input, every output — automation is usually the right tool. If the system has to decide what to do next based on what it finds, that is when an agent earns its keep. Most SMEs do not need agents for most of their workflows. But when they do, the difference in outcome is significant."
Freelance SEO & AI Consultant — London
Automation or agent — how to choose
The right tool depends on whether the workflow is predictable or adaptive. Both are valid — the mistake is applying the wrong one.
Use when the workflow is fully predictable
- Every step can be mapped in advance
- Inputs are consistent and structured
- The system follows a fixed trigger-action sequence
- No decision-making required between steps
- Zapier, Make, or n8n handles it without LLM reasoning
Use when the system has to decide what to do next
- Inputs vary and the system must interpret them
- Multiple tools or data sources need to be orchestrated
- The workflow branches based on reasoning, not fixed rules
- The agent needs to handle exceptions without human intervention
- The process cannot be fully mapped before it runs
What the implementation engagement involves
Four phases from scoping to deployment — with architecture design as the critical first step.
Scoping and architecture design
We define the agent's scope, the tools it will use, the data sources it will access, and the decision logic it will apply. Agent architecture is designed before any code is written — the most common failure mode is building before scoping.
Build and integration
The agent is built and integrated with your existing systems — CRM, data sources, communication tools, or internal APIs. Integration is the step that determines whether the agent is useful in practice or only in isolation.
Testing against edge cases
Agents are tested against the full range of inputs they will actually receive — including edge cases and failure modes. An agent that works on clean inputs but fails on messy real-world data is not production-ready.
Deployment and maintenance plan
The agent is deployed with monitoring in place and a maintenance plan that defines what triggers a review. Agents degrade when the business changes — the maintenance plan is what keeps them useful over time.
Phase 01 (scoping and architecture) is highlighted because the most common failure mode is building before scoping — an agent built around a vague brief will require significant rework once it meets real-world inputs.
Agent use cases for London SMEs
Four categories where agent-level reasoning consistently produces a measurable return — because they replace work that is too complex for automation but too high-volume for humans.
Research and synthesis agents
Agents that gather information from multiple sources, synthesise it into a structured output, and flag exceptions for human review. Replaces high-volume, low-complexity research work that currently consumes disproportionate senior time.
Triage and routing agents
Agents that receive unstructured inputs (emails, enquiries, support requests), classify them, extract key information, and route them to the right person or system. Particularly effective for high-volume inbound workflows.
Multi-step process agents
Agents that handle workflows spanning multiple tools and decision points — for example, receiving a brief, checking availability, drafting a response, and logging the interaction — without human intervention at each step.
Monitoring and alerting agents
Agents that watch for specific conditions across data sources and trigger actions when those conditions are met. Replaces manual monitoring work and reduces the lag between a signal appearing and a response being made.
Scoped per engagement before work begins. Price reflects the complexity of the agent, the number of integrations, and the maintenance requirements. No hourly billing, no scope creep.
Not sure whether a custom agent is the right solution? The £999 assessment identifies the highest-value AI opportunities and the right implementation tier for each one.
If the workflow is predictable and can be fully mapped in advance, automation is usually the better choice — faster to build, easier to maintain, and lower cost.
If the need is information retrieval and on-brand output rather than multi-step reasoning, a custom GPT and knowledge system is the right level of complexity.
AI implementation and SEO work together — a retainer ensures the organic channel keeps pace with the operational improvements the agent delivers.
Frequently asked questions
What is the difference between an AI agent and automation?
Automation handles predictable workflows — every step can be mapped in advance, inputs are consistent, and the system follows a fixed trigger-action sequence. An agent handles work that requires reasoning and adaptation — the system has to decide what to do next based on changing inputs, orchestrate multiple tools, and handle exceptions without a fixed script. If the process can be fully mapped before it runs, automation is usually the better choice. If the system needs to reason about what to do next, that is when an agent earns its keep.
Do I need to have completed the AI Tools Assessment first?
For an engagement at this price point, a scoping session is always required before work begins — whether that is a standalone discovery call or the full AI Tools Assessment (C16). The assessment is the more thorough option: it maps all your workflows, identifies the highest-value opportunities, and produces a prioritised implementation roadmap. If you already have a specific agent use case in mind and want to scope it directly, a discovery call is the right starting point.
What does 'SME-appropriate' mean in practice?
It means the agent is scoped for the complexity and budget of a small or medium business — not enterprise architecture with enterprise costs. In practice: we build what is needed, not what is technically possible; we use tools and platforms that your team can maintain without specialist support; and we scope the engagement so the agent is useful within weeks, not months. The £5–10K price range reflects this — it is not a fixed fee, it is a range that scales with the actual complexity of what needs to be built.
What happens after the agent is built?
The agent is deployed with monitoring in place and a maintenance plan that defines what triggers a review. Agents degrade when the business changes — a new data source, a changed process, a different input format — and the maintenance plan is what keeps them useful over time. Simple agents can be maintained internally. More complex agents with multiple integrations are typically reviewed on a quarterly basis, either independently or as part of an ongoing retainer arrangement.
How is this different from hiring a developer?
A developer builds what you specify. This engagement starts from the business problem, not the technical specification — the scoping and architecture phase is where we determine what needs to be built, not just how to build it. The deliverable is a working agent that solves a specific business problem, with documentation and a maintenance plan, not a codebase that requires ongoing developer support to operate.
Build an agent that earns its keep
Book a 30-minute discovery call. We will review your use case, confirm whether an agent is the right solution, and scope the build before any commitment is made.