peterbamuhigire AI-Agent Vertical Positioning
Acknowledgement: Shared by Peter Bamuhigire, techguypeter.com, +256 784 464178.
Use When
- The AI-agent bid is in a defined vertical and the proposal must demonstrate vertical literacy in agentic terms.
- The buyer's evaluators include vertical specialists (regulator, sector champion, clinical lead, head of supervision, bar council, ombudsman).
- The competitive set includes horizontal vendors who do not understand the vertical's autonomy stance.
- The vertical carries sector-specific autonomy limits (financial-services decisioning, healthcare clinical separation, legal practice rules, public-sector administrative law).
Do Not Use When
- The engagement is horizontal and the vertical is not a discriminator.
- The vertical is non-agent (use
ai-on-saas-vertical-positioning).
Domain Inputs
- The buyer's vertical and sub-vertical.
- The regulator(s) and stance on autonomous decisioning.
- The candidate agent use cases.
- The competitive set (named or inferred).
- The buyer's irreversibility tolerance and HITL stance.
- Local-content / sovereign requirements.
Domain Method
- Identify the vertical and load the matching vertical reference (
vertical-ai-agent-<vertical>.md). - Choose two to four named agent use cases appropriate to the vertical, each with autonomy level and reversibility profile.
- Position the agency's discriminators for the vertical: regulator literacy, autonomy discipline, irreversibility gating, kill-switch readiness, action audit log, redress mechanism, sovereign-AI option, language coverage, supervisor retraining capability.
- Frame risk and regulator language using the vertical's vocabulary, not generic agent language (FS: model-risk management and adverse-decisioning rights; healthcare: clinical decision support classification and non-diagnostic agentic; legal: bar rules and lawyer responsibility; public sector: administrative-law fairness and sovereign-AI).
- Choose the win themes from ai-agent-win-themes-and-discriminators that fit the vertical.
- Draft the vertical positioning paragraph for Executive Summary, the vertical case studies for Relevant Experience, and the vertical methodology variations for
06-methodology.
Vertical Library (Brief)
Customer Support — Resolution Agents
- Agent use cases: ticket triage and routing; deflection on FAQ and account self-serve; agent-assist (suggest reply, draft macro, find article); end-to-end resolution on bounded action sets (refund up to value bound, ticket close, status update); voice-of-customer summarisation.
- Autonomy stance: L3 on triage; L2 on agent-assist; L1 on irreversible (refund, cancellation, account closure).
- Regulator: data-protection regulators (general); sectoral where the support is for regulated services.
- Discriminators: containment-vs-deflection honesty (containment is bad if the customer left the journey frustrated); agent-assist before agent-replace framing; escalation SLA with named human; multilingual depth; supervisor retraining curriculum.
Financial Services — Compliance and Reconciliation Agents
- Agent use cases: KYC document triage and exception assembly; AML alert disposition assembly (the agent assembles the case, a named human dispositions); transaction reconciliation across systems; complaint classification and routing; sanctions-screening triage.
- Autonomy stance: L3 on triage and case assembly; L1 (strict) on disposition, closure, and any customer-facing action.
- Regulator: CBK, CBN, SARB, BoU, NBR, FSCA, NDPC; model-risk management discipline; adverse-decisioning rights; FATF guidance on AI in AML; SR 11-7-style governance for groups with US footprint.
- Discriminators: model-risk overlay on every agent; HITL strict on decisioning; the agent never approves / declines / closes alerts on its own; full audit log with regulator-grade access; sovereign-AI option; reconciliation determinism with replay.
Insurance — Claims Triage Agents
- Agent use cases: claims-document triage and categorisation; fraud-signal assembly (the agent assembles the signal pack, a human decides); first-notification-of-loss summarisation; policy-comparison drafting for customer service; subrogation candidate identification.
- Autonomy stance: L3 on triage and assembly; L1 on declinature, payout, fraud determination.
- Regulator: IRA (KE / UG), NAICOM, FSCA; conduct-of-business; fairness in claims handling.
- Discriminators: explainability of triage; HITL strict on declinature and payout; citation grounding; complaint-channel awareness; sectoral retraining for claims handlers shifting from doer to supervisor.
Public Sector — Citizen-Service Agents (Extreme Caution)
- Agent use cases: citizen-FAQ and information retrieval (cited); case routing inside ministries; multilingual translation of forms; complaint triage and routing; document understanding for procurement / judiciary; non-decisioning support to caseworkers.
- Autonomy stance: L1 / L2 only on anything affecting a citizen's right or entitlement. L0 on benefit determinations, immigration decisions, criminal-justice decisions, tax assessments. The agent assembles; the named officer decides.
- Regulator: NCAIS Kenya, NAIS Nigeria, NITA-U Uganda, RGB / Rwanda AI Policy, ICASA / DCDT South Africa; administrative-law fairness; right to human review; transparency-to-affected-party; sovereign-AI expectations.
- Discriminators: sovereign-AI option (on-prem or in-country inference); local-language depth; administrative-law literacy (no decisioning); transparency-to-affected-party UX with contestability; published action catalogue; quarterly kill-switch drill with regulator observer.
- Honesty principle — the proposal does not promise citizen-service agents that take decisions on citizens. Public-sector buyers reward this honesty; competitors who over-promise lose to enforcement.
Healthcare — Admin-Only Agents
- Agent use cases: clinical-documentation drafting (summary, discharge letter, referral) with clinician sign-off; prior-authorisation triage; claims and billing reconciliation; appointment scheduling and follow-up; hospital-operations triage.
- Autonomy stance: L2 / L3 on admin; L0 on anything clinical — diagnosis, treatment recommendation, dose decision, triage of patient acuity for life-critical care.
- Regulator: Ministries of Health, KMPDC, NHIS / SHA / SHIF, HPCSA; HIPAA where US-linked; clinical-decision-support classification and the line where AI becomes a medical device.
- Discriminators: explicit non-clinical scope on the action catalogue; clinician-in-the-loop on every drafted clinical artefact; audit log with retention matched to clinical record-keeping rules; residency.
Legal — Drafting and Review Agents
- Agent use cases: contract first-draft generation and clause comparison; due-diligence document review (the agent flags; the lawyer decides); legal-research synthesis with citations; filing assembly (the agent assembles; the lawyer files); e-discovery triage; client-intake drafting.
- Autonomy stance: L2 / L3 on drafting and assembly; L0 on filings, advice, and any communication purporting to be from a lawyer or to bind a client. The lawyer remains the responsible party; bar rules require it.
- Regulator: Bar associations and law societies (Uganda Law Society, Law Society of Kenya, NBA, GCB, GCB, SRA in UK-aligned jurisdictions); civil-procedure rules; confidentiality and privilege rules.
- Discriminators: confidentiality by tenant scope (no shared memory across firms); citation grounding (no hallucinated cases); audit log; lawyer-final on every filing and communication; insurance arrangement aligned to E&O.
Operations — Alerting, Triage, and Coding Agents
- Agent use cases: alert triage (incident routing, runbook execution on bounded actions); coding agents (refactor, test generation, dependency update, pull-request authorship under human review); infrastructure-change triage (the agent proposes; the SRE applies); log-summary and root-cause assembly.
- Autonomy stance: L2 / L3 on triage and assembly; L1 on production changes; L2 / L3 on reversible code changes within a sandbox; L1 on merge-to-main and any prod deployment.
- Regulator: data-protection (general); sectoral where the operations are inside a regulated service.
- Discriminators: deterministic replay; sandbox by default; merge-gate human review; observability of the agent itself; rollback by default; coding-agent specific eval (test pass, regression, dependency-safety, secret-leak).
Sample Vertical Paragraphs (illustrative)
Financial Services
Our financial-services agentic practice treats every agent as a model under model-risk management. Agents triage and assemble; named officers decide. The agent never approves or declines a credit, an AML alert, or a sanctions match on its own. The action catalogue is published, every action class has a reversibility classification, and irreversible actions are L1-gated with a named approver. The audit log is regulator-grade and replay-deterministic so an internal auditor or a supervisor can reconstruct any decision the agent influenced. A sovereign-AI option is available for the most sensitive tenants.
Public Sector
Our public-sector agent practice begins with what an agent must not do. Agents do not decide on benefits, status, tax, or justice; agents assemble the case, in the language the citizen speaks, with citations to the controlling regulation, and a named officer decides. The action catalogue is published. Every external communication discloses the agent and offers an immediate route to a human. Sovereign-AI deployment is supported for ministries whose data must not leave the country. The kill-switch drill is run quarterly, with a regulator observer where invited.
Healthcare
Our healthcare agent practice is explicitly admin-only. Agents draft discharge summaries, route prior-authorisations, reconcile claims, schedule follow-ups; agents do not diagnose, do not triage clinical acuity for life-critical care, and do not recommend treatment. Every drafted clinical artefact is signed by the responsible clinician. The action catalogue makes the non-clinical boundary explicit and auditable, and the methodology classifies any expansion across that boundary as a regulated-device decision that pauses the engagement until the regulator concurs.
Quality Standards
- Vertical use cases are named and bounded; no generic "agents for the sector".
- Autonomy stance is per use case, with reversibility profile.
- Regulator references are specific.
- Discriminators are vertical-specific.
- Risk framing uses the vertical's vocabulary.
- Honesty principle is applied — promises the regulator would not accept are not made.
- Case studies or comparable references are in the vertical or transferable with stated rationale.
Domain Risks
- "Agents for banking" with no named use case.
- L4 / L5 autonomy promised for citizen-service or healthcare clinical decisioning.
- Legal drafting agents promised without lawyer-final on filings.
- Coding agents that auto-merge to main without human review.
- Insurance claims agents that decide payout or declinature on their own.
- Customer-support agents that close tickets without resolution evidence (containment masquerading as deflection).
- Operations agents that act on production without a sandbox or merge gate.
Domain Outputs
- Vertical positioning paragraph for Executive Summary.
- Vertical-specific use-case list with autonomy and reversibility profile.
- Vertical discriminators block.
- Vertical risk and regulator language.
- Vertical methodology variations for
06-methodology. - Vertical case studies or comparable references.
Anti-Patterns
- Inventing a metric, credential, constraint, or buyer position. Fix: cite the supplied source or mark the item as an assumption requiring confirmation.
- Treating an unavailable check as passed. Fix: mark it not assessed and state the evidence needed to resume.
- Advancing autonomy without a named gate owner. Fix: require observable evidence, accountable acceptance, and a rollback path.
- Reusing another sector or use case without reassessment. Fix: retest affected parties, action scope, reversibility, and jurisdiction.
- Writing acceptance as “satisfactory” or “appropriate”. Fix: define an observable measure, threshold, evidence record, and decision owner.
Inputs
| Artefact | Source/provider | Required? | Missing-input behaviour |
|---|---|---|---|
| target sector, buyer problem, jurisdiction, affected parties, proof, and autonomy constraints | Buyer evidence, ToR, approved discovery record, system owner, or measured operating data | Yes | Stop the affected decision; list the missing source and return only a qualified outline or assumption register. |
Outputs
| Artefact | Consumer | Acceptance condition |
|---|---|---|
| Sector-positioned proposal narrative | Buyer evaluator and account lead | Scope, assumptions, exclusions, owners, decision logic, and observable acceptance tests are explicit and traceable to supplied evidence. |
Evidence Produced
| Evidence | Consumer | Acceptance condition |
|---|---|---|
| sector-positioned proposal narrative | Buyer evaluator and account lead | Every load-bearing claim traces to supplied evidence; assumptions, owners, gates, exclusions, and observable acceptance conditions are explicit. |
Capability Contract
Default to read-only for discovery, analysis, review, and planning. Minimum capability is access to the supplied artefacts and permission to calculate or inspect evidence. Edit only the requested proposal working copy. Do not change production systems, contact affected parties, publish, spend, certify compliance, or approve autonomous action without explicit authority from the accountable owner.
Degraded Mode
If files, interviews, telemetry, specialist review, network access, or calculation tools are unavailable, produce the narrowest useful qualified result. Mark each unavailable check as not assessed, separate facts from assumptions, lower confidence, and state the evidence needed to resume. An unassessed gate is never a pass.
Decision Rules
| Choice | Action | Failure or risk avoided |
|---|---|---|
| Select sector stance and use case | Use verified sector evidence and set autonomy by consequence, reversibility, and regulatory exposure. | Generic positioning or unsafe cross-sector reuse. |
| Required evidence, authority, or accountable owner is missing | Stop the affected recommendation or commitment and record the gap. | Invented evidence or unauthorised autonomy. |
| Gate evidence is complete and accepted | Advance only within the approved scope and retain the evidence trace. | Scope drift and irreproducible approval. |
Workflow
- Confirm the consumer, authority, neighbouring-skill route, and required inputs; stop when a mandatory source or accountable owner is missing.
- Inspect the evidence and record facts, assumptions, conflicts, and unavailable checks; stop on a failed safety, finance, regulatory, or acceptance gate.
- Apply the domain method and decision rules within the qualified scope, retaining an evidence trace.
- Draft the contracted output and reconcile it with methodology, work plan, staffing, pricing, risk, and governance; recover by revising the affected scope or control and rerunning the failed gate.
- Verify acceptance conditions, permission boundaries, direct references, and anti-slop controls; block release until failed checks are corrected.
Worked Example
For healthcare, position an administrative scheduling agent with human review and no clinical decision authority; do not reuse the customer-support autonomy stance without a fresh risk assessment.
References
- vertical-ai-agent-customer-support
- vertical-ai-agent-financial-services
- vertical-ai-agent-insurance
- vertical-ai-agent-public-sector
- vertical-ai-agent-healthcare
- vertical-ai-agent-legal
- vertical-ai-agent-operations
- ai-agent-win-themes-and-discriminators
- ai-on-saas-vertical-positioning — AI-on-SaaS verticals (load alongside for SaaS-embedded agents).
- ai-agent-methodology — methodology that variations sit inside.
- ai-agent-risk-and-responsible-ai — vertical-specific risk entries.