Service-as-Software — Pricing Models for AI Agents That Replace SaaS

Research date: 2026-03-19 | Agent: Deep Research | Confidence: High

Executive Summary

The “SaaSpocalypse” is real but nuanced — software stocks lost ~$2 trillion in early 2026, but global SaaS spending is still projected to reach $512B by 2028; the model is transforming, not dying
Foundation Capital’s $4.6T “service-as-software” thesis is now the default B2B AI investing frame — AI delivers outcomes, not tools, targeting the services market (4x larger than software)
Outcome-based pricing is the fastest-growing model — Intercom Fin ($0.99/resolution), Zendesk ($1.50–$2.00/resolution), and Sierra (custom) prove it works; Gartner projects 40% of SaaS contracts will include outcome-based components by 2026
AI agent startups are hitting unprecedented growth — Cursor $1B ARR in 24 months, Sierra $100M in 21 months, Devin from $1M to $73M in 9 months
For Moklabs, the hybrid model is optimal — base platform fee + usage/outcome tiers, aligning spend with value while maintaining revenue predictability

Market Size & Growth

The “Service-as-Software” Opportunity

Metric	Value	Source
Global services market (target)	$4.6 trillion	Foundation Capital
Global SaaS spending (2025)	$318 billion	Intellectia
Global SaaS spending (2028 est.)	$512 billion	Intellectia
Global SaaS spending (2029 est.)	$576 billion	Intellectia
Software market cap lost (Feb 2026)	~$2 trillion	TechCrunch
AI agents market (2025)	$7.6–7.8 billion	AI Funding Tracker
AI agents market (2026 est.)	$10.9 billion	AI Funding Tracker
AI agents market (2030 est.)	$52.6 billion	AI Funding Tracker
AI agents CAGR	46.3%	Salesmate
Enterprise apps embedding agents by 2026	40% (up from <5% in 2025)	Gartner via Salesmate
SaaS contracts with outcome-based components by 2026	40%	Gartner via Chargebee

Fastest-Growing AI Agent Companies

Company	Domain	ARR	Growth Rate	Valuation	Funding
Cursor (Anysphere)	Coding	$1B+ (Nov 2025)	$100M → $1B in 10 months	—	—
Harvey	Legal	$195M (end 2025)	3.9x YoY	$8B → $11B (Feb 2026)	$960M+ total
Sierra	Customer Service	$100M (21 months)	—	~$10B	—
Glean	Enterprise Search	$200M+	Doubled in 9 months	$7.2B	$150M Series F
Cognition (Devin)	Coding	$73M (Jun 2025)	$1M → $73M in 9 months	—	—
ElevenLabs	Voice AI	$330M+ (end 2025)	3x valuation in months	$11B	$500M Series D

Key Players & Pricing Models

The Eight Pricing Structures

Model	How It Works	Best For	Example
Per-Seat	License per user (human or AI “seat”)	Stable teams, predictable usage	Traditional SaaS, some AI wrappers
Per-Agent	License per deployed AI agent	Multi-agent orchestration	OctantOS potential model
Usage-Based	Pay per token, API call, or compute	Developer tools, APIs	OpenAI, Anthropic
Per-Workflow	Pay per automated workflow execution	Process automation	Make, Zapier
Per-Output	Pay per generated artifact	Content/code generation	Image generators
Outcome-Based	Pay per successful result	Customer service, sales	Intercom Fin, Sierra, Zendesk
Subscription (Flat)	Fixed monthly/annual fee	All-you-can-eat usage	ChatGPT Plus
Hybrid	Base subscription + usage/outcome tiers	Most AI-native companies (recommended)	Devin, Relevance AI

Real-World Pricing Deep Dives

Intercom Fin — The Outcome-Based Pioneer

Price: $0.99 per resolution (successful outcome)
How it works: You pay only when Fin fully resolves a customer issue without human intervention
Minimum: 50 resolutions/month
Standalone: Fin can be purchased without the Intercom platform at $29/mo
Key insight: Charged once per conversation, even if Fin resolves multiple questions
Source: Intercom

Zendesk AI — Dynamic Budget Remix

Price: $1.50/resolution (committed) or $2.00/resolution (pay-as-you-go)
Innovation: Dynamic Pricing Plan lets you shift budget between human seats and AI resolutions during your contract — no renegotiation needed
Example: Start at 20% automation → ramp to 60% → shift budget accordingly
Key insight: Bridges the transition from per-seat to per-resolution without forcing customers to choose one model
Source: Zendesk

Sierra AI — Custom Outcome-Based

Price: Custom per-resolution pricing (quote-based, ~$150K+ annually)
Model: Gets paid only when completing a task for the customer
Flexibility: Blended pricing for different interaction types — outcome-based for resolutions, consumption-based for routing/greeting
Growth: $100M ARR in 21 months proves the model works
Source: Sierra

Devin (Cognition) — Agent Compute Units

Tiers: Core ($20/mo), Teams ($500/mo), Enterprise (custom)
Unit: Agent Compute Unit (ACU) — ~15 min of active work
Price per ACU: $2.25 (Core) / $2.00 (Teams, 250 included)
Key insight: Normalized unit that abstracts away token costs, VM time, and inference — customers pay for “work time” not infrastructure
Growth: $1M → $73M ARR in 9 months
Source: Devin

Harvey — Premium Seat-Based for Legal

Price: $1,000–$1,200/lawyer/month, 25–50+ seat minimums
Annual contracts: $30K–$300K+
Growth: Median seat count doubles within 12 months of deployment
Key insight: Seat-based works in regulated, high-value verticals where per-outcome measurement is complex
Source: Sacra

Glean — Enterprise AI Search

Price: ~$45–50+/user/month base, +$15/user/month for AI add-ons
Minimum: $50K–$60K annual contract (100+ seats)
PoC cost: Up to $70K
Key insight: Enterprise-first, quote-based — high ASP justified by “find anything in your company” value prop
Source: GoSearch

Technology Landscape

The “Service-as-Software” Thesis Explained

Traditional model (Software-as-a-Service):

Customer buys a tool → Customer is responsible for using it → Customer achieves outcome (or doesn’t)

New model (Service-as-Software):

Customer describes desired outcome → AI agent delivers the outcome → Customer pays for result

The key shift: responsibility for achieving the outcome moves from the customer to the vendor. Instead of QuickBooks (tool), you get an AI accountant (service). Instead of Zendesk (tool), you get Sierra (service).

Why This Matters for Pricing

Factor	SaaS Pricing	Service-as-Software Pricing
Value unit	Seat / user access	Outcome / completed work
Revenue scales with	Headcount growth	Work volume / automation rate
Customer risk	High (pay even if unused)	Low (pay only for results)
Vendor risk	Low (predictable revenue)	Higher (must deliver outcomes)
Margin model	70–85% gross margin	Variable (depends on cost per outcome)
Expansion motion	Add seats	Increase automation scope
Churn risk	High if underutilized	Low (tied to business value)

The Three Value Axes for Pricing Decisions

Per Chargebee:

Value Attribution — How easily can customers tie agent outputs to outcomes? (High = outcome-based pricing viable)
Execution Autonomy — How well does the agent solve problems without human-in-loop? (High = per-resolution pricing viable)
Workload Predictability — How spiky is the effort per instance? (Low = usage-based better than flat-rate)

The Bessemer Hybrid Recommendation

Bessemer Venture Partners recommends hybrid models for most AI companies:

Base subscription ensures revenue predictability for both vendor and customer
Usage/outcome tiers capture upside as customer value grows
Best for: Uncertain or early markets, quantifiable verticals (legal, support, sales)
Real example: EvenUp and Legora (legal AI) use hybrid pricing in quantifiable verticals where AI output ties to specific outcomes

Pain Points & Gaps

For AI Agent Vendors

Margin uncertainty with outcome-based pricing — Any work consumed by the agent in attempting a solution goes under-monetized when only successful outcomes are charged. Failed attempts still cost tokens.
Definition of “resolution” is contentious — What counts as a successful resolution? Intercom charges when the customer doesn’t request more help; Zendesk requires full resolution without human intervention. Different definitions → different economics.
Cost-of-goods-sold unpredictability — Complex queries might cost 10x more tokens than simple ones, but the charge is the same per resolution. Without robust cost attribution (see Agent Economics report), margins are opaque.
Usage-based pricing surprises — 78% of IT leaders reported unexpected charges on SaaS due to consumption-based or AI pricing models (per Zylo). Trust erosion is a real risk.

For Enterprise Buyers

Budget unpredictability — Hard to forecast costs when pricing varies with automation rate and resolution volume
Vendor lock-in concerns — Outcome-based models often require deep integration, making switching costly
Measuring true ROI — If one AI seat replaces 5 human seats but costs 3x a human seat, is that good? Depends on quality, reliability, and edge cases.
“AI-washing” in pricing — Some vendors just add a chatbot overlay and call it “AI-powered” while charging premium prices

Underserved Segments

SMBs with <50 employees — Most AI agent platforms require $50K+ annual contracts. The $20–50/mo tier is thin.
Platform operators (like Moklabs) — Need pricing infrastructure (metering, billing) built into the orchestration layer, not bolted on
Vertical-specific pricing — Generic models don’t capture domain-specific value (legal resolution vs. support resolution vs. code task)

Opportunities for Moklabs

1. Built-in Usage Metering & Billing in Paperclip (High Impact / High Strategic Value)

Current state: Paperclip tracks agent budgets and costs but has no customer-facing billing/metering.

Opportunity: If OctantOS customers deploy agents for their end users, they need to meter and bill for agent work. Building metering primitives into Paperclip means:

Track ACUs (Agent Compute Units) per task/issue
Generate usage reports per customer/project
Support configurable pricing models (per-task, per-outcome, hybrid)
Export to billing systems (Stripe, Orb, Flexprice)

Why it matters: This is the “Stripe for AI agents” play — enabling OctantOS customers to monetize their agent deployments with built-in pricing infrastructure.

Estimated effort: 4–6 weeks for core metering, 2–3 months for full billing integration

2. Hybrid Pricing for OctantOS (High Impact / Medium Effort)

Recommended pricing structure (building on MOKA-57 proposal):

Tier	Base Fee	Included	Overage
Free	$0	3 agents, 100 tasks/mo	N/A
Pro	$39/user/mo	10 agents, 1,000 tasks/mo	$0.10/task
Enterprise	$999/mo + $25/user	Unlimited agents, 10,000 tasks/mo	$0.05/task + volume discounts

Key design decisions:

Base fee anchors revenue predictability (Bessemer recommendation)
Per-task overage captures expansion without surprising customers
Task-based (not token-based) pricing shields customers from model cost volatility
Free tier drives adoption; conversion from free → pro at 100 tasks/mo limit

3. Outcome-Based Pricing Option for Vertical Deployments (Medium Impact / High Effort)

Opportunity: For specific verticals where outcomes are measurable (customer support, code generation, document processing), offer outcome-based pricing as an option:

Per-resolution for support agents
Per-completed-task for coding agents
Per-processed-document for document agents

Why it matters: Zendesk’s “dynamic pricing plan” proves that customers want to shift between seat-based and outcome-based during their contract. Building this flexibility into OctantOS would be a differentiator.

4. Pricing Analytics Dashboard (Medium Impact / Low Effort)

Opportunity: Help OctantOS customers understand their cost-per-outcome and optimize:

Cost per resolution/task/outcome breakdown
Agent efficiency metrics (cost vs. value delivered)
Pricing model comparison tool (“would outcome-based save you money?”)
Margin analysis for service-as-software businesses

5. “Devin-Style” ACU Abstraction for Agent Work (High Impact / Medium Effort)

Opportunity: Define an “Agent Work Unit” (AWU) for Paperclip that normalizes:

LLM token costs across providers
Compute time for tool execution
Storage and context costs

Customers think in “work done” not “tokens consumed.” This abstraction enables:

Predictable per-task pricing regardless of which models are used
Model routing optimizations invisible to the customer
Clean billing that non-technical buyers can understand

Risk Assessment

Market Risks

Risk	Likelihood	Impact	Mitigation
Outcome-based pricing fails to scale (too many failed attempts eat margins)	Medium	High	Hybrid model with base fee ensures minimum revenue
Race to bottom on per-resolution pricing ($0.99 → $0.49 → $0.10)	Medium	Medium	Compete on platform capabilities, not price per resolution
Enterprise buyers resist variable pricing (prefer predictable budgets)	Medium	Medium	Offer committed volume discounts (Zendesk model)
SaaS incumbents successfully add AI agents without changing pricing	High	Medium	Move fast — incumbents are slow to cannibalize per-seat revenue

Technical Risks

Risk	Likelihood	Impact	Mitigation
Defining “outcome” is domain-specific and complex	High	Medium	Start with clear definitions (task completed, issue resolved)
Usage metering at scale adds latency and complexity	Medium	Low	Async metering with eventual consistency
Multi-model cost normalization is hard	Medium	Medium	Start simple: cost = tokens × model_price, iterate

Business Risks

Risk	Likelihood	Impact	Mitigation
Moklabs too early for monetization (need users first)	High	High	Free tier first, pricing infrastructure as foundation for later
Building billing competes with core product work	Medium	High	Start with metering only — billing integrates with Stripe/Orb later
Pricing complexity confuses early customers	Medium	Medium	Default to simple hybrid, advanced options for enterprise

Data Points & Numbers

Pricing Benchmarks by Category

Category	Company	Model	Price Point
Customer Service	Intercom Fin	Per resolution	$0.99/resolution
Customer Service	Zendesk AI	Per resolution	$1.50–$2.00/resolution
Customer Service	Sierra	Per resolution (custom)	~$150K+/year
Customer Service	Ada	Per resolution	Quote-based
Legal	Harvey	Per seat (premium)	$1,000–$1,200/lawyer/mo
Legal	EvenUp	Hybrid	Base + outcome
Coding	Devin	Per ACU	$2.00–$2.25/ACU (~15 min work)
Coding	Cursor	Subscription + usage	$20/mo (Pro), $200/mo (Ultra)
Enterprise Search	Glean	Per seat (premium)	$45–50+/user/mo
Platform/API	OpenAI	Per token	$2.50–$15/M input tokens
Platform/API	Anthropic	Per token	$3–$15/M input tokens

Key Financial Metrics

Metric	Value	Source
IT leaders with unexpected AI charges	78%	Zylo
Klarna AI assistant impact	Replaced ~700 agents	Multimodal
AI agents ROI within year 1	74% of enterprises	Agility at Scale
Operational cost reduction with AI	75% average	Agility at Scale
Customer willingness to pay for AI outcomes	Growing — 40% contracts outcome-based by 2026	Gartner

The SaaSpocalypse Numbers

Metric	Value	Source
Software market cap lost (Feb 2026)	~$2 trillion	TechCrunch
Market cap erased in 7 days	$1 trillion+	Digital Applied
AI-native startups launched (18 months)	Thousands	Foundation Capital
Global services market (AI target)	$4.6 trillion	Foundation Capital