Workflow Agents: done for you.
An agent executes a multi-step workflow autonomously; a human sets the goal and reviews. A function gets faster and cheaper.
Three companies, three targets. Each moves to the level its P&L needs.Three portfolio companies, three targets. What ships in one transfers to the next.
- Month-end close, Industrials: moves from Level 1 to Level 2. 14-day close, all by hand. 10-day close, numbers trusted.
- Insurance approvals, Healthcare: moves from Level 1 to Level 3. 45 min per patient, staff on the phone. 20 min, chatbot finds each insurer's rules. 4 min by agent, nurse signs off · est. $310K/yr.
- Renewal quoting, Software: moves from Level 1 to Level 4. 2 days per quote. Same day. 1 hour, agent drafts. Minutes · est. $480K/yr.
- Month-end close, Industrials portco: moves from Level 1 to Level 2. 14-day close at the portco. 10 days; the fund sees numbers a week sooner.
- Insurance approvals, Healthcare portco: moves from Level 1 to Level 3. 45 min per patient, staff on the phone. 20 min, chatbot finds each insurer's rules. 4 min by agent, nurse signs off · est. $310K/yr to EBITDA.
- Renewal quoting, Software portco: moves from Level 1 to Level 4. 2 days per quote. Same day. 1 hour, agent drafts. Minutes · est. $480K/yr · playbook to the next portco.
Three companies, three targets. Each moves to the level its P&L needs.Three portfolio companies, three targets. What ships in one transfers to the next.
- Month-end close, Industrials: moves from Level 1 to Level 2. 14-day close. 10-day close.
- Insurance approvals, Healthcare: moves from Level 1 to Level 3. 45 min each. 20 min each. 4 min · $310K/yr.
- Renewal quoting, Software: moves from Level 1 to Level 4. 2 days. Same day. 1 hour. Minutes · $480K.
- Month-end close, Industrials portco: moves from Level 1 to Level 2. 14-day close. 10-day close.
- Insurance approvals, Healthcare portco: moves from Level 1 to Level 3. 45 min each. 20 min each. 4 min · $310K/yr.
- Renewal quoting, Software portco: moves from Level 1 to Level 4. 2 days. Same day. 1 hour. Minutes · $480K.
Azure
Software runs a whole step; a person checks the edge cases.
Who does the workSoftware, you review
A human sets the goal and reviews; the agent picks steps, calls tools, retries, and produces the deliverable.
One targeted pilot (the single highest-ROI workflow, shipped production-grade in 100 days) proves the economics and wins the adoption argument. Every level is reached the same way: one workflow at a time.
Same workflow, different question: who does the work.
Level 3: Workflow Agents
An agent runs the close end to end. People review the exceptions it flags and approve.
Cost per seat becomes cost per task. This is the line between a faster headcount and a cheaper function.
Cost per seat becomes cost per task
Below L3, a copilot makes a person faster and is priced per person using it. At L3 the agent runs the step and a person reviews the edge cases, so the cost is per unit of work done. That is what moves EBITDA.
A function gets faster and cheaper
- Finance: variance, narrative and board-deck agents, chained. FP&A 10 days to 2.
- Support: the agent drafts the reply; a human reviews. Deflection moves from 30% toward 70%+.
- Engineering: Claude Code in CI opens PRs from failing tests and lint debt. PR throughput 2× vs L1.
How L3 stalls
- Agentifying an unmeasured process: no baseline cycle time or error rate, so you can't tell if the agent helped.
- Write access without rollback. The agent sends 400 wrong-segment emails. Trust dies for 18 months.
- Tool soup: three agent platforms, no shared evals, no shared identity.
- Time for L4 when you want agents to call other agents, and compliance asks for replay, audit and lineage.
Moving from L2 to L3 takes a total company investment of $750K to $1.75M, over 9 to 12 months. That range covers the data foundation, platform and run cost underneath the agent, not only our fees.
The named picks at L3.
L3 is where the pilot proves the economics. Build drops a production-grade agent into your stack in 100 days: finance variance, sales SDR, support tier-1, AP, recruiting, product research. Each ships with the governance we build into every agent: audit trails, per-agent cost attribution, evals, and access controls, running in your own cloud account.
What L3 actually looks like in a portfolio company
| Function | In practice | Signal | What it saves |
|---|---|---|---|
| Finance | Variance, narrative and board-deck agents, chained, evaluated, observable | FP&A 10d → 2d | 7 to 8 days per FP&A cycle |
| Sales | The agent enriches, scores and routes leads; humans close | Conversion uplift | |
| Support | Agent drafts the reply; human reviews; ticket closes faster with audit trail | AI-first triage | Deflection from 30% toward 70%+ |
| Legal | Agent feeds the deal desk a risk-scored summary | Not a 60-page redline | |
| Engineering | Claude Code / Cursor Agent in CI opens PRs from failing tests & lint debt | PR throughput 2× vs L1 |
Six Build blueprints.
Variance / FP&A agent
Close-to-commentary cycle compressed 10d → 2-3d. Variance agent reads the close package, drafts commentary, flags exceptions, routes to the CFO for sign-off.
SDR agent
Research → personalization → send → reply classification → meeting. Pipeline doesn't grow because reps work harder; it grows because the agent works while reps sleep.
Tier-1 agent
Agent drafts the reply; human reviews. Deflection moves from 30% (L2) toward 70%+, without losing the long tail that needed a human.
AP agent
Invoice intake → matching → approval routing → exception escalation. Pairs with Ramp / AppZen where appropriate.
Recruiting agent
Inbound screening → scheduling → first-round scorecard. The recruiter takes the shortlisted candidates into final rounds, not the inbox triage.
Product Research agent
The agent watches usage signals, drafts feature briefs, summarizes user research, builds the PM the agenda for the weekly review. PM ships, doesn't curate.
How L3 stalls.
- Agentifying an unmeasured process. No baseline cycle time or error rate; can't tell if the agent helped.
- Cursor agent on autopilot, no observability. Token bills explode; devs disable it.
- Write access without rollback. Agent sends 400 wrong-segment emails. Trust dies for 18 months.
- Zapier-tax surprise. Great agent + 100K tasks/mo = $1,500+/mo with no migration plan.
- Tool soup. Three agent platforms, no shared evals, no shared MCP server, no shared identity.
- Picking Agentforce because the SI prefers it. Lock-in is real; agent math is opaque.
What L4 looks like.
- The workflow has natural specialization (research / draft / review / route / escalate).
- Persistent state across runs is a hard requirement.
- You want agents to call other agents.
- Compliance asks for replay, audit, lineage. You need governed orchestration.
- Cost-per-task replaces cost-per-seat as the metric.
Common questions.
What is Level 3 AI maturity?
Level 3 (L3) is Workflow Agents: software runs a whole step and a person checks the edge cases. An agent executes a multi-step workflow autonomously; a human sets the goal and reviews. The agent picks steps, calls tools, retries, and produces the deliverable.
What changes at Level 3?
Cost per seat becomes cost per task. Below Level 3, a copilot makes a person faster and is priced per person using it. At L3 the agent runs the step and a person reviews the edge cases, so the cost is per unit of work done.
What does a Level 3 agent look like in practice?
A function gets faster and cheaper. Finance: variance, narrative and board-deck agents, chained, taking FP&A from 10 days to 2. Support: the agent drafts the reply and a human reviews, with deflection moving from 30% toward 70%+. Engineering: Claude Code in CI opens PRs from failing tests, with PR throughput 2x versus Level 1.
How long does it take to reach Level 3?
One targeted pilot, the single highest-ROI workflow, ships production-grade in 100 days and proves the economics. The full move from Level 2 to Level 3 runs 9 to 12 months. Every level is reached the same way: one workflow at a time.
How much does it cost to get from Level 2 to Level 3?
Moving from Level 2 to Level 3 takes a total company investment of $750K to $1.75M, over 9 to 12 months. That range covers the data foundation, platform and run cost underneath the agent, not only our fees.
How are Level 3 agents kept safe and governed?
A person reviews the edge cases, and every agent we ship carries audit trails, per-agent cost attribution, evals, and access controls. The failure to avoid is write access without rollback: the agent sends 400 wrong-segment emails and trust dies for 18 months. So we baseline cycle time and error rate before an agent touches the process.