The 8-primitive substrate (Pillar P1) doesn't change much across verticals. What changes is which agents, which tasks, which knowledge, which regulators, which integrations. This essay walks through four verticals — accounting, legal, customer support, marketing — and shows the playbook each one runs.
The vertical-specific framing matters because the agentic-firm pattern only sells when it's expressed in the vocabulary of the buyer's domain. "We've built an agent platform" is a non-sale. "Here's how a 12-person accounting practice runs the close on agents" is a sale.
Vertical 1 — Agentic Accounting
Who buys: practice owners + CFO-as-a-Service firms doing bookkeeping, monthly close, tax prep, advisory.
Pain: seasonal capacity crunch (close periods, tax season) + low-margin commodity work + senior partner time drained by mechanical work.
The agent fleet (typical):
- Bookkeeping agent — categorises transactions, flags anomalies. Heart: webhook-triggered on QuickBooks/Xero transaction posts.
- Close-prep agent — runs the month-end checklist, reconciles accounts, escalates breaks. Heart: schedule-triggered first business day of each month.
- Tax-research agent — answers tax-treatment questions from senior practitioners with citations to current IRC sections. Heart: manual + A2A from senior workspace.
- Advisory-content agent — drafts client-facing memos from underlying data. Heart: A2A from a client-engagement agent.
Regulator surface: AICPA + PCAOB (US), ICAEW (UK), local equivalents. PCAOB's AI position is evolving — agent decisions still require attribution to a CPA, but the evidence artefact requirements are softening.
Key citations: IRS Modernized e-File, AICPA SAS 145, Big-4 advisory whitepapers on AI in audit.
Live AgentsBooks satellites: agentic-accounting-firm (long-form), accounting-agent-glossary (terms), finance-blueprints (downloadable workflows).
Vertical 2 — Agentic Legal Practice
Who buys: boutique law firms, in-house legal ops, contract-management practices.
Pain: contract review backlog + discovery cost + due-diligence quality variance.
The agent fleet:
- Contract-review agent — first-pass redline against firm playbook + risk-tier classification. Heart: webhook on inbound contract email.
- Discovery agent — runs document-set queries against case-relevant predicates with citation back to the source document. Heart: A2A from senior litigation agent.
- Due-diligence agent — runs the standard DD checklist against a target-firm dataset, flags red flags. Heart: schedule + manual.
- Client-update agent — drafts weekly client updates from the case-management system. Heart: schedule (Fridays).
Regulator surface: ABA (US), SRA (UK), state bar associations. ABA's Formal Opinion 512 on generative AI (2024) sets the duty of competence + confidentiality framework — agents are explicitly permitted with supervisor sign-off.
Key citations: ABA Model Rules 1.1 + 1.6, LegalSifter + Harvey public case studies, Stanford CodeX research on AI in legal practice.
Live AgentsBooks satellites: agentic-legal-practice.
Vertical 3 — Agentic Customer Support
Who buys: B2C scale-ups, B2B SaaS support orgs, marketplace customer experience teams.
Pain: ticket volume scales with growth; quality drops at scale; senior CX time drained by tier-1 work.
The agent fleet:
- Tier-1 resolver — handles the routine ~70% of tickets end-to-end. Heart: webhook on inbound ticket.
- Tier-2 escalator — when tier-1 isn't confident, routes to the right specialist agent or human. Heart: A2A from tier-1.
- Trend-watcher — surveys recent tickets for emerging issues (e.g., "spike in payment failures from EU users on iOS 18"). Heart: scheduled hourly.
- VOC synthesizer — synthesises voice-of-customer themes for the product team. Heart: scheduled weekly.
Regulator surface: lighter — CCPA/GDPR for data handling, FTC for marketing/consumer-protection content. The strict regimes are in HC + financial-services subsegments.
Key citations: Intercom Fin metrics, Klarna newsroom, Zendesk benchmark reports.
Live AgentsBooks satellites: agentic-customer-support, support-agents-compared, case-study-intercom-fin.
Vertical 4 — Agentic Marketing Team
Who buys: content-led B2B companies + DTC brands + content agencies.
Pain: content volume per channel keeps climbing; quality is inconsistent; founder/CMO time spent on operations rather than strategy.
The agent fleet:
- Content-research agent — runs the topic-research sprint (R1..R8 dimensions per our research-cadence pattern). Heart: triggered by topic-pick.
- Content-drafting agent — drafts long-form essays / spokes / newsletter from the research notes. Heart: A2A from research-agent on sprint complete.
- Cross-post agent — adapts published essays for dev.to, Medium, Hashnode, LinkedIn — each with the canonical link pointing home. Heart: webhook on publish.
- Comment / DM responder — handles inbound social engagement under a defined voice profile. Heart: event-triggered.
- Performance analyst — daily report on what's working, which channels, which topics. Heart: scheduled.
Regulator surface: FTC (US, endorsement/disclosure), CAP Code (UK), GDPR/CCPA for any personalisation. Disclosure that content is AI-assisted is a customer-trust matter even when not legally required.
Key citations: Stack Overflow developer survey, social media benchmark reports from Hootsuite / Sprout Social, Andrew Chen's distribution writing.
Live AgentsBooks satellites: agentic-marketing-team, marketing-blueprints, sdr-agent-glossary.
What's common across the four
All four verticals share five design choices that fall out of the substrate:
- Each agent has explicit Identity (named, role-tagged, owner-attributed).
- Each Heart task carries a budget cap — sustainability lives at the task level, not the firm level.
- Each task emits structured audit artefacts (intent + evidence + decision + confidence).
- Senior practitioners review escalations, not routine output.
- The eval harness runs continuously — without it, every model change is a risk; with it, model changes are a regression test.
The vertical playbooks differ in content but converge on shape. That's the value of the substrate.
Frequently asked questions
Q: Which vertical should an AgentsBooks customer start with?
A: Whichever they're already in. The 4 verticals above all have a free-tier starter on AgentsBooks; the firm-starter for that vertical is a one-click clone.
Q: What about verticals not listed (HR, recruiting, real-estate, insurance)?
A: Same pattern works; we have early customers in each. The 8-primitive substrate is vertical-agnostic; the playbook is what you build on top.
Q: How does this map to the 8 primitives?
A: Each agent in the fleet uses all 8 — the playbooks are how the substrate manifests in a specific domain.
Pick your vertical, clone the firm starter: Start free →
