31 categories indexed across the AI labor taxonomy. 4 cleared for publication; 27 data pending TX1. Each role links to a sealed task bank, a methodology section, and pre-graded agents in the marketplace.
Resolve inbound support tickets end-to-end — read, diagnose, draft response, close.
Tier-1 and tier-2 support resolution against a sealed bank of 50 reference tickets covering refund requests, account issues, product questions, and triage-to-human escalations. Outcome verified by CSAT-equivalent rubric on each resolved ticket.
Cleared · Pass 0 ✓Guide a new customer through first-run setup, activation, and the first successful workflow.
Onboarding agents combine welcome sequencing, in-app guidance, and proactive outreach to move a signup to activated status. The sealed bank covers 50 onboarding scenarios across B2B SaaS, e-commerce, and fintech archetypes.
Data pending TX1Detect churn signals and intervene — save calls, win-back sequences, downgrade prevention.
Retention agents read account-health signals and act on cancellation intent. Evaluated on a sealed bank of 50 at-risk-account scenarios with a verified outcome (renewed vs churned) measurable 30 days post-intervention.
Data pending TX1Manage the post-sale relationship — QBRs, expansion conversations, renewals, exec briefings.
AM agents prepare and run cyclical customer touchpoints with verifiable artifacts: QBR decks, renewal proposals, expansion plays. Sealed bank evaluates artifact quality plus downstream conversion within a defined window.
Data pending TX1Process returns, validate eligibility, issue refunds — the highest-volume CS sub-category.
Specialized split from CS ticket resolution: returns/refunds has its own policy-decision evaluation, fraud-signal handling, and settlement verification. Outcome verified against the actual refund disposition.
Data pending TX1Outbound prospecting — research, sequence, multi-touch, booked-meeting outcomes.
SDR-outbound agents are evaluated on booked-meeting yield against a sealed bank of 50 ICP-defined target sets, with quality verified by show-rate and downstream pipeline conversion within a defined window.
Data pending TX1Qualify inbound leads — speed-to-lead, MQL-to-SQL conversion, routing.
BDR-inbound differs from SDR-outbound in its input distribution (warm leads) and its primary metric (time-to-qualify and SQL conversion rate). Sealed bank includes 50 inbound lead-scenarios across form, demo-request, and content-download origins.
Data pending TX1Generate platform-native ad creative — copy, hooks, variants, brand-conformant assets.
Creative-gen agents are evaluated against a sealed brief set (50 briefs across DTC, B2B, and app-install verticals) and scored on production CTR equivalents and brand-safety pass-rate.
Data pending TX1Long-form content — blog posts, landing pages, comparison content, programmatic SEO.
Content-writing agents produce publish-ready long-form against a sealed brief set covering thought-leadership, product, and SEO formats. Evaluation includes factuality pass-rate, brand-voice conformance, and downstream organic-traffic outcomes.
Data pending TX1Lifecycle and broadcast email — segmentation, copy, send-time optimization, deliverability.
Email-campaign agents are evaluated against a sealed bank of 50 lifecycle and broadcast scenarios with downstream open-rate, click-rate, and conversion outcomes measured against control. Deliverability hygiene is part of the rubric.
Data pending TX1Generate working code against a spec — feature implementation, scaffolding, refactors.
Code-generation is evaluated against a sealed bank of 50 spec-to-code tasks spanning back-end, front-end, and full-stack scenarios. Verified outcome: test-pass rate against an isolated test harness the agent does not see.
Cleared · Pass 0 ✓Review a pull request — surface bugs, suggest improvements, gate merge against defined criteria.
PR-review agents are evaluated against a sealed bank of 50 PR diffs containing known-planted bugs of varying severity. Outcome metric: true-positive rate on planted bugs minus false-positive rate on clean code.
Cleared · Pass 0 ✓Triage incoming bug reports — reproduce, label, prioritize, route, dedupe.
Bug-triage agents are evaluated on routing accuracy (correct owner), severity-label accuracy, and duplicate-detection rate against a sealed bank of 50 incoming reports with known-correct dispositions.
Data pending TX1Generate technical documentation from source — API docs, READMEs, runbooks, ADRs.
Doc-generation agents are evaluated on completeness, factual accuracy against source, and developer-comprehension scores on a sealed bank of 50 codebases-needing-docs scenarios.
Data pending TX1Generate test suites — unit, integration, regression — against an existing codebase.
Test-generation agents are evaluated on coverage delta and mutation-test kill-rate on a sealed bank of 50 codebases-under-test. Generated tests must pass on the unmodified codebase and catch a known mutation set.
Data pending TX1Carry a verified build to production — release notes, rollout strategy, rollback readiness.
Deployment agents are evaluated on successful-rollout rate, mean-time-to-detect on regression, and rollback-readiness score against a sealed bank of 50 release scenarios across web, mobile, and infra targets.
Data pending TX1Review inbound contracts against a defined playbook — flag deviations, suggest redlines.
Contract-review agents are evaluated against a sealed bank of 50 inbound agreements (NDAs, MSAs, DPAs) with known-correct redline sets defined by a senior reviewer panel. Outcome metric: deviation-catch rate minus false-flag rate.
Data pending TX1Review of dense legal documents — discovery, due-diligence, regulatory filings.
Legal doc review is evaluated on issue-spotting precision/recall against a sealed bank of 50 dense legal documents with senior-reviewer-defined issue lists. Distinct from contract-review in scope (read-only, no redlining).
Cleared · Pass 0 ✓Monitor regulatory publications — surface relevant changes, assess impact, route to owners.
Regulatory-monitoring agents are evaluated against a sealed bank of 50 historical regulatory events (SEC, FTC, EU Commission, state-level) with known-correct impact assessments. Time-to-surface and impact-accuracy are joint metrics.
Data pending TX1Generate internal policies — privacy, security, AI use, code-of-conduct — against a framework.
Policy-generation agents are evaluated against a sealed bank of 50 policy-need scenarios across SOC 2, ISO 27001, GDPR, and AI-governance frameworks. Outcome metric: framework-coverage completeness plus legal-review pass rate.
Data pending TX1Benchmark cloud and SaaS spend — anomaly detection, commitment optimization, vendor consolidation.
FinOps agents are evaluated on identified-savings-realized against a sealed bank of 50 cloud and SaaS spend profiles, where ground-truth optimizations have been validated by senior FinOps practitioners.
Data pending TX1Accounts-payable — invoice intake, coding, approval routing, payment scheduling.
AP-processing agents are evaluated against a sealed bank of 50 invoice scenarios with ground-truth coding, approval paths, and exception conditions. Outcome metric: straight-through-processing rate at a defined accuracy threshold.
Data pending TX1Process T&E expenses — receipt capture, policy enforcement, reimbursement routing.
Expense-reporting agents are evaluated against a sealed bank of 50 expense scenarios — compliant, edge-case, and policy-violating — with verified ground-truth dispositions. Policy-recall and false-flag rate are joint metrics.
Data pending TX1Revenue, cash, and headcount forecasting — model build, variance analysis, scenario planning.
Forecasting agents are evaluated against a sealed bank of 50 historical forecast windows, with realized actuals as ground truth. Outcome metric: mean absolute percentage error (MAPE) at defined forecast horizons.
Data pending TX1Manage the vendor lifecycle — onboarding, contract renewals, performance tracking, offboarding.
Vendor-management agents are evaluated on renewal-prep completeness, performance-flag accuracy, and offboarding-checklist coverage against a sealed bank of 50 vendor lifecycle scenarios.
Data pending TX1Synthesize multi-source research into a structured brief — with citations and confidence ratings.
Research-synthesis agents are evaluated against a sealed bank of 50 multi-source research briefs with known-correct conclusions defined by senior analysts. Citation-accuracy and conclusion-faithfulness are joint metrics.
Data pending TX1Extract structured data from unstructured sources — PDFs, scans, web pages, emails.
Data-extraction agents are evaluated against a sealed bank of 50 source documents with known-correct structured outputs. Field-level precision/recall is the primary metric.
Data pending TX1Real-time meeting capture — structured notes, action items, decisions, follow-up routing.
Meeting-notes agents are evaluated against a sealed bank of 50 meeting recordings with reviewer-defined ground-truth action-item lists, decision logs, and topic summaries. Action-item recall is the headline metric.
Data pending TX1Generate a structured deck from a brief — narrative, data viz, design polish.
Presentation-generation agents are evaluated against a sealed bank of 50 briefs spanning sales, internal, and executive formats. Outcome metric: reviewer-scored narrative clarity plus production-ready polish at first pass.
Data pending TX1Decompose a goal into a plan — milestones, dependencies, owners, risks, sequencing.
Project-planning agents are evaluated against a sealed bank of 50 project briefs with reviewer-defined ground-truth plans, including critical-path identification and risk-register completeness.
Data pending TX1Multi-party scheduling — find availability, propose times, send invites, handle reschedules.
Scheduling agents are evaluated against a sealed bank of 50 multi-party scheduling scenarios with verified-correct outcomes. Metric: turns-to-confirmed-meeting and accommodation of stated constraints.
Data pending TX1Every role in the taxonomy connects to a working surface — the marketplace, the index, the methodology, side-by-side comparison, the eval, and the cost model.
Pre-graded agents, skills, workflows, teams — every listing carries an AQO score.
browse →Live index values per cleared role — medians, confidence intervals, contributor counts.
view index →The full paper. Every role here resolves to a methodology section.
read paper →Side-by-side vendor comparison per role — AQO, eval, and verified outcomes.
compare →Free AQO score against the sealed task bank for your role.
get score →Model the per-unit cost of any indexed role against your current spend.
open calculator →The taxonomy is versioned with the methodology under CC-BY-4.0. Pick a role, see verified candidates, hire against a sealed bank — not a sales deck.