The previous guide defined what an agent is; this one is about where they actually pay. The patterns below recur across service firms because they share a shape: frequent, rule-guided, spread across systems, and important enough to hurt when done late or badly, which is exactly the work that never gets a dedicated hire. These are working patterns, not concepts; variants of most of them run inside my own business daily.
Triage: deciding what deserves attention
Everything that arrives, enquiries, referrals, documents, support requests, needs the same first work: read it, classify it, route it, acknowledge it. An agent does this continuously and consistently: urgency assessed, the right owner notified, the CRM updated, a holding reply drafted for approval. The Kick-starter page shows this exact workflow as its worked example, because it is so often the right first agent: high frequency, bounded actions, obvious measurement.
Chasing: the follow-ups nobody enjoys
Missing documents from a client. Unpaid invoices. Unsigned forms. Renewal dates. Chasing is pure agent territory: the state lives in your systems, the rules are knowable, the drafting is formulaic, and the cost of it slipping is real money and real risk. Deployed, it looks like: agent watches the state, drafts each chase in your voice with the correct details, a person approves the send, escalations surface on schedule. The human gate sits at the send, until trust is earned for the routine tier.
Reconciliation: keeping systems agreeing
Two systems that should match and quietly do not: the ledger and the bank feed, the CRM and the back office, the fee expected and the fee received. Agents read both sides, match what matches, and present only the exceptions with their evidence, turning an afternoon of comparison into ten minutes of judgement. This pattern rewards agents precisely because it is boring: attention fails at row 400, agents do not.
Drafting pipelines: documents from known inputs
Where a document’s content is determined by inputs you already hold, suitability letters from a fact-find, reports from a data pack, meeting summaries from a transcript, an agent assembles the draft, flags what is missing, and routes it for review. The person’s hour moves from typing to judging, which is where their value was all along. The suitability-drafting playbook covers the advice-firm version in depth.
Monitoring and research: watching so people don’t have to
Sources that matter to your firm, regulator updates, market moves, client-sector news, competitor changes, watched continuously, with only the items meeting your brief surfaced, summarised and cited. This is the pattern behind this site’s own content operations, productised in the Managed Websites service: an agent-run editorial desk with a human on the approval gate. The same machinery pointed inward becomes a briefing service for your own team.
The constants across every pattern
Reading the five again, the same design repeats, and it is the whole discipline in miniature:
- Narrow job, written down
- Tool access matched to the job, no wider
- Agent does the continuous legwork; a person holds the gate where consequences live
- Every action logged
- Measurement from day one
- Widen only on evidence
Skip any of those and you meet the failure statistics; keep them and the governance guide becomes a checklist you already satisfy.
The five patterns against the usual first-agent test:
| Pattern | Frequency | Consequences gated? | Typical first agent? |
|---|---|---|---|
| Triage | Continuous | Yes: routing and drafts | The classic starter |
| Chasing | Daily-weekly | Yes: approval on sends | Strong starter |
| Reconciliation | Daily-monthly | Internal by nature | Strong starter |
| Drafting pipelines | Per matter | Yes: human reviews | Second wave, high value |
| Monitoring & research | Continuous | Internal by nature | Easy win, softer savings |
Where I fit in
Matching these patterns to your firm’s actual processes is precisely what an Automation Audit produces: your recurring work mapped, ranked by effort against impact, with the right first agent identified. Building it is a Kick-starter; running a whole operation this way is the Software Factory. Bring the process your team complains about most, and I will tell you honestly which pattern it is and what it would take.