Klarna's AI agents handled 2.3 million customer service conversations in their first month — the equivalent output of 700 full-time human agents — and resolved tickets in under 2 minutes versus an 11-minute human average.
That is not task automation. That is workflow replacement. And it is the clearest signal that enterprise AI has crossed a threshold most companies have not fully processed yet.
Task AI vs. Workflow AI: The Distinction That Determines Valuation
Most enterprise AI deployments in 2023–2024 were task-level: a Copilot that helps draft an email, a summarizer that condenses a meeting, a classifier that routes a ticket to the right queue. Useful. Not transformative.
Agentic AI operates at the workflow level. It owns the entire process — intake, decision-making, execution, and output — without a human checkpoint at each step. The commercial difference is not marginal:
| Dimension | Task AI | Workflow AI (Agentic) |
|---|---|---|
| Human involvement | Per step | Exception-only |
| Value delivered | Faster individual work | Headcount displacement |
| ARR multiple (2025–2026) | 10–15x | 40–80x |
| Churn risk | High (feature parity) | Low (deep workflow embed) |
| ROI timeline | Days to weeks | Weeks to months |
| Enterprise buyer | Dept. manager | CFO / COO |
The Workflows Going Agentic First
Not all workflows are equally ready for agentic AI. The ones moving fastest share three traits: high transaction volume, semi-structured data inputs, and measurable SLA-based outcomes. Here is where the penetration is real:
Klarna: 700 FTE equivalent in month one, 2-minute average resolution vs. 11-minute human average. Sierra (customer service AI) raised at $4.5B valuation. The ROI case is the clearest of any vertical.
Devin autonomously completes engineering tasks end-to-end — debugging, feature building, test writing — with a $2B valuation on ~$20M ARR. Copilot Workspace orchestrates multi-file changes across a full PR lifecycle. Dev agents are now running in production at 40%+ of Fortune 500 engineering orgs.
Harvey handles entire matter workflows — discovery review, brief drafting, contract negotiation redlines — not just clause extraction. EvenUp automates personal injury demand letters end-to-end. Harvey raised at $1.5B with $50M+ ARR as of Q1 2026.
ServiceNow reports AI-assisted workflows resolving 85%+ of tier-1 support tickets without human escalation at major enterprise clients. Moveworks autonomous resolution rates exceed 60% across its installed base. The IT helpdesk headcount model is being repriced.
Spend categorization, anomaly detection, AP workflow routing, and audit trail generation are being handled autonomously. Ramp reports 80%+ of expense reports processed without human review at customers above 500 seats.
What "Agentic AI Enterprise" Means for Valuation
The multiple gap between task AI and workflow AI is not subtle. Per CB Insights and PitchBook data from 2025–2026 private rounds:
10–15x ARR
Point-solution AI (task-level)
20–35x ARR
Horizontal AI platform (workflow-adjacent)
40–80x ARR
Agentic AI (full workflow ownership)
Salesforce Agentforce illustrates why: it closed 5,000+ enterprise deals in its first six months (Q4 2024 through Q1 2025), positioning Salesforce as the default enterprise agent orchestration layer across its existing CRM install base. The company that owns the workflow owns the renewal.
The Headcount Question Enterprises Are Not Asking Clearly Enough
Klarna did not lay off 700 people when they deployed AI agents. They froze hiring and let attrition compress headcount. That is how this plays out — not a dramatic moment, but a structural shift in how enterprises grow.
Gartner projects that by 2028, 15% of day-to-day work decisions will be made autonomously by AI agents. McKinsey Digital puts the number of workflows fully automatable by AI agents at 30%+ by 2026 — up from roughly 7% in 2023. The distinction matters: full workflow automation is not the same as productivity augmentation. One reduces headcount requirements. The other increases output per person.
For investors, the signal is clear: companies selling workflow AI to enterprise buyers are in a fundamentally different business than those selling productivity AI. The former competes on ROI measured in eliminated FTEs. The latter competes on features. One of those is stickier, higher ACV, and less vulnerable to model commoditization.
What to Watch in the Next 18 Months
Agent orchestration layers consolidating
Microsoft, Salesforce, ServiceNow competing for the enterprise agent OS — the winner controls the API layer where agents run
Vertical agents IPO-ing or getting acquired
Harvey, Sierra, Cognition are the most likely candidates for acquisition by Salesforce, Oracle, or ServiceNow at $3B–$10B
Agent liability frameworks emerging
First major enterprise AI error with financial consequences will trigger regulatory and contractual frameworks around agent accountability
Human-in-the-loop requirements in regulated industries
Financial services and healthcare will mandate human review thresholds, creating a ceiling on workflow AI penetration in those sectors through 2027
The enterprise AI market is not splitting between "AI haves" and "AI have-nots."
It is splitting between companies that own a workflow and those that just help with a step inside one.
Track enterprise AI adoption and agentic AI deployment trends on the Enterprise AI Adoption Dashboard and AI Valuations Dashboard at Value Add VC.