Autonomous Operations: anticipatory AI.
Level 5 is the system as colleague: it initiates work, governed by policies, not prompts. L5 is the frontier, and we say so.
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
The system notices work that needs doing and starts it.
Who does the workSoftware, governed by policy
L5 is emerging. We don't pretend otherwise.
Real L5 is largely aspirational in 2026. Here's what actually exists today:
| What exists today | What it actually is |
|---|---|
| L4+ cron | Scheduled agents that look proactive but are trigger-reactive. |
| L4+ memory | Durable memory layers (Mem0, LangMem, Letta/MemGPT) bolted onto L4. |
| Narrow L5 · coding | Devin parallel sessions, Cursor background agents, and Claude Code in CI proactively opening PRs from failing tests or lagging deps. |
| Narrow L5 · CX | Sierra and Decagon's newest tier, paid on resolution. |
| Bespoke internal | Top engineering orgs (Anthropic, OpenAI, Stripe, Shopify), pattern proven, not productized. |
| "Marketing L5" | L3/L4 stacks branded "Operator," "agentic OS," or "autonomous employee", mostly L4 with a memory layer and a content team. |
The defensible L5 question: What does it do when nobody is logged in?
The frontier, and we say so.
Level 5: Autonomous Operations
The system starts the close when the ledger locks, and asks for a person only when policy says so.
The frontier, and we say so. Almost nobody operates here yet.
Where L5 exists today
- Engineering. Parallel autonomous SWE agents shipping features overnight. The most credible L5 production reference today.
- Support. Tier 1, tier 2, escalation triage: all agent-run. Sierra and Decagon are the production references in 2026.
- Finance. Close, reconciliation, variance, commentary: driven by agents. Human approval only at board sign-off.
- Product-embedded. The core experience is a conversation with an agent that runs the job the user hired the product to do.
How L5 marketing differs from L5 reality
- Calling cron L5. A scheduled agent is L4 with a timer.
- L5 demos that need an audience. If it only runs when leadership is watching, it isn't autonomous.
- No kill switch, no scope. Autonomy without enforced policy boundaries becomes incidents.
- Skipping L3 to L4. No leapfrog. Teams that try produce demos, not systems.
Why L5 moves the multiple
- L3 is where the economics compound. L4 is where they inflect. L5 is where exit multiple moves, and where our work becomes R&D rather than implementation.
- For software companies, an agent-native product is a category change, not a feature add. Pricing shifts from per seat to per outcome, and the buyer is pricing a different business.
- For non-software portcos, L5 governance is the exit asset. The buyer sees a fund that already classified durability risk before they did.
L5 is not yet banded, and the estimator stops at L4. We co-build with portfolio companies that want to be early to the operating environment shift: the work is R&D-flavored, priced per engagement, and we're honest about what's solved and what's still being invented.
The named picks at L5.
At L5, our value is being an early, sophisticated co-builder. The work is R&D-flavored, not implementation-flavored. We're not inventing the engine; we build and govern the operating environment around it (cost, evals and audit) so a CFO can sign a contract.
What L5 looks like in the rare places it exists.
Close-to-report by agents
Close, reconciliation, variance, commentary: driven by agents. Human approval only at board sign-off. The CFO reviews exceptions; the agents run the cycle.
Full-funnel agent organization
Research to negotiation support, as a fleet. The seller closes; the agents do everything else. Pricing model shifts from seat-based to outcome-based.
Full resolution stack
Tier 1, tier 2, escalation triage: all agent-run. Sierra and Decagon are the production references in 2026.
Sourcing to offer modeling
Talent acquisition as an agent organization. The talent leader signs offers and manages culture; the fleet does the rest.
Devin fleet on the backlog
Parallel autonomous SWE agents shipping features overnight. Most credible L5 production reference category today.
Agent-native product
The core experience is a conversation with an agent (or agents) that runs the job the user hired the product to do. Pricing shifts seat → outcome. The exit multiple reflects a category shift.
How L5 marketing differs from L5 reality.
- Calling cron L5. A scheduled agent is L4 with a timer.
- L5 demos that need an audience. If it only runs when leadership is watching, it isn't autonomous.
- No kill-switch, no scope. Autonomy without enforced policy boundaries becomes incidents.
- Memory without governance. Long-lived memory is a long-lived liability.
- Skipping L3 to L4. No leapfrog. Teams that try produce demos, not systems.
Why L5 moves the multiple.
- "L3 is where the economics compound. L4 is where they inflect. L5 is where exit multiple moves, and where the partnership becomes R&D rather than implementation."
- For software companies, an agent-native product is a category change, not a feature add. Pricing model shifts from per-seat to per-outcome, and the buyer is pricing a different business.
- For non-software portcos, L5 governance is the exit asset. The buyer sees a fund that already classified durability risk before they did.
Common questions.
What is Level 5 AI maturity?
Level 5 (L5) is the system as colleague: it notices work that needs doing and starts it, governed by policies, not prompts. The work is done by software, governed by policy. L5 is the frontier, and we say so.
Is Level 5 AI real today?
Real Level 5 (L5, where an operation runs itself under human oversight) is largely aspirational in 2026. What exists is narrow: parallel autonomous coding agents shipping features overnight, the most credible production reference today, and support tiers from Sierra and Decagon paid on resolution. Much of what is sold as L5 is L4 with a memory layer and a content team.
How much does it cost to reach Level 5?
Level 5 (L5, where an operation runs itself under human oversight) is not yet banded, and the estimator stops at L4. We co-build with portfolio companies that want to be early to the operating environment shift: the work is R&D-flavored, priced per engagement, and we're honest about what's solved and what's still being invented.
How do I tell real Level 5 from marketing?
Ask what it does when nobody is logged in. A scheduled agent is Level 4 (L4, where several agents coordinate across functions) with a timer; a demo that only runs when leadership is watching isn't autonomous; and autonomy without a kill switch and enforced policy boundaries becomes incidents.
Can a company skip straight to Level 5?
No. There is no leapfrog over Level 3 (L3, where software runs a whole workflow and a person reviews the exceptions) and L4; teams that try produce demos, not systems. L3 is where the economics compound, L4 is where they inflect, and L5 is where the exit multiple moves, and where our work becomes R&D rather than implementation.
Why does Level 5 matter at exit?
For software companies, an agent-native product is a category change, not a feature add: pricing shifts from per seat to per outcome, and the buyer is pricing a different business. For non-software portfolio companies, Level 5 (L5, where an operation runs itself under human oversight) governance is the exit asset: the buyer sees a fund that already classified durability risk before they did.