What is service-as-software?

Service-as-software is the business model where AI agents complete entire job functions and customers pay per outcome (resolved ticket, sourced candidate, closed contract) instead of per seat. Bret Taylor popularized the term in 2024 and Foundation Capital codified the thesis. It contrasts with SaaS, which charges humans to operate software, by charging companies for the work itself.

How is service-as-software different from regular SaaS?

SaaS sells software that humans use; service-as-software sells the output of work that AI agents do. SaaS charges per seat and assumes a human is in the loop. Service-as-software charges per result and the human only audits. The pricing flips from headcount-denominated to throughput-denominated, which is why it appears on enterprise procurement as a payroll line, not a software line.

Which SaaS categories are most exposed to service-as-software replacement in 2026?

The eight categories with the highest replacement velocity right now are customer support, sales outreach, recruiting and sourcing, legal research and drafting, marketing operations, IT helpdesk, finance and accounting, and customer success. Each has well-defined outputs and largely deterministic workflows, the two conditions service-as-software requires to win.

Will service-as-software replace all SaaS by 2030?

No. Service-as-software wins for deterministic, single-output workflows but does not win for live synchronous collaboration, strategic decision-making, or open-ended creative work. Tools for visual collaboration, real-time meetings, and shared decision-making are likely to grow as more individual-contributor work moves to agents. The right framing is "service-as-software eats workflow tools; collaboration tools become more important."

How should I evaluate a service-as-software vendor?

Compare cost per outcome against your current per-seat total cost of ownership, not against a competitor's per-seat list price. Audit the agent's accuracy on a 100-task representative sample before signing. Check that pricing scales sub-linearly with volume so you actually capture savings. And demand auditability — if you can't see why an agent made each decision, you're buying a black box and a future compliance problem.

Service-as-Software: 8 SaaS Categories AI Agents Replace

# Service-as-Software: 8 SaaS Categories AI Agents Replace

Bret Taylor's Sierra raised $950M at a $15.8B valuation this week. Decagon, Glean, and Harvey each crossed $1B. Foundation Capital's service-as-software thesis — that AI agents will not assist software users but replace them — is no longer a venture deck. It is the funding pattern of 2026.

The shift matters because it inverts the SaaS pricing model. For 25 years, vendors charged per seat for software that humans operate. Service-as-software charges per outcome — per resolved ticket, per sourced candidate, per closed deal — for an agent that does the job. Your SaaS bill stops being a software cost and starts being a payroll line.

This guide breaks down the 8 SaaS categories where the service-as-software model is already winning, the one cluster where it is quietly failing, and what your stack should look like after the dust settles in 2026.

What Service-as-Software Actually Means

"Service-as-software" is the term Foundation Capital coined and Bret Taylor popularized to describe AI agents that complete entire job functions instead of helping a human do them faster. The classic SaaS model packaged workflows as software interfaces. The service-as-software model packages the workflow's output as a billable service. The customer never opens a UI.

Three forces made this viable in 2026, not 2024:

Reliability passed the threshold. Anthropic, OpenAI, and Google reasoning models now solve domain-specific tasks at 90%+ accuracy with proper scaffolding — the bar enterprises require to retire human-in-the-loop QA.
Agent commerce protocols arrived. Cloudflare and Stripe's agent payments protocol and Visa's Intelligent Commerce launched May 2026, giving agents authenticated identity and the ability to transact without a human session.
Buyers ran out of patience for SaaS sprawl. Average enterprises now run 106 SaaS apps with only 49% license utilization, per Gartner. A service-as-software vendor that replaces 4 SaaS tools with one outcome contract is an easy budget conversation.

That trifecta is why $4.6B flowed into agentic startups in Q1 2026 alone, more than the total in 2024. Below are the categories absorbing that capital — and pulling SaaS revenue with it.

1. Customer Support: Sierra, Decagon, Intercom Fin

Customer support was the first category service-as-software won, and it is the clearest blueprint. Sierra's agents resolve 70%+ of tickets autonomously and bill per resolved conversation. Klarna replaced 700 contact-center FTEs with an OpenAI-based agent and reports $40M annual savings. Decagon and Ada now sign enterprise deals in the eight-figure range.

The SaaS that loses: per-seat support desks (Zendesk's stagnant growth is no accident), QA scoring tools, and the lower tier of BPO contracts. The remaining human role is escalation manager, not first-line agent.

2. Sales Outreach: 11x, Artisan, AISDR

The SDR was the textbook entry-level white-collar role. In 2026 it is being replaced wholesale. 11x's "Alice" agents prospect, write personalized sequences, handle objection-handling email threads, and book meetings — billed at a flat monthly fee per agent-FTE-equivalent. Artisan's billboards ("Stop hiring humans") made the category infamous.

The SaaS that loses: outbound platforms (Outreach, Salesloft), enrichment tools (ZoomInfo, Apollo), and sequencing software priced per seat. Conversion data still favors human-AI hybrids for enterprise deals, but for transactional SMB outbound, the agentic stack already wins on cost per meeting booked.

3. Recruiting and Sourcing: Mercor, Paradox, Hirevue Agents

Mercor reached a $2B valuation by replacing the sourcing function: agents read every résumé, score against job criteria, run a video screen, and forward only finalists. Paradox's "Olivia" runs the high-volume hourly hiring funnel for McDonald's, Wendy's, and CVS — millions of candidates per year, zero recruiters in the first three stages.

The SaaS that loses: per-recruiter ATS seats, sourcing tools (LinkedIn Recruiter at $10,800/year per seat), and the rec-coordinator role itself. The service-as-software contract is "$X per qualified candidate delivered," not "$Y per recruiter seat."

4. Legal Research and Drafting: Harvey, Legora, Spellbook

Harvey hit a $5B valuation in March 2026 by selling outcomes — drafted contracts, due diligence memos, deposition prep — instead of legal research seats. Legora raised a Series D extension from Atlassian Ventures this month. Spellbook automates contract redlining at $99/contract instead of $499/seat.

The SaaS that loses: per-lawyer research platforms (Westlaw, LexisNexis are repricing fast), contract lifecycle management tools, and the second-year associate review cycle. AmLaw 100 firms are quietly cutting recruiting classes by 15-30%, citing agentic productivity gains.

5. Marketing Operations: Jasper Agents, Copy.ai Workflows, AdCreative.ai

The marketing ops function — campaign briefs, asset generation, A/B testing, performance reporting — is being absorbed by agentic platforms that bill per campaign launched. Jasper's new "Marketing Operating System" runs the entire workflow from brief to creative to deployment, signing seven-figure ARR with Fortune 500s.

The SaaS that loses: campaign management seats (Marketo, Pardot pricing pressure), creative production tools sold per editor, and the freelance marketplace tail. A 12-person marketing team is now operationally equivalent to a 40-person team from 2023.

6. IT Helpdesk and Service Management: Moveworks, Aisera, Glean Work AI

Moveworks (acquired by ServiceNow for $2.85B in March 2025) and Aisera resolve 60-80% of L1/L2 IT tickets without a human ever touching them: password resets, software access requests, onboarding provisioning, MDM enrollment. Glean Work AI extends the model to internal knowledge — agents answer "how do I file expenses?" with the canonical company answer, automatically.

The SaaS that loses: per-seat IT service management tools, internal knowledge base products, and the L1 helpdesk role entirely. Buyers are reporting 40-60% reductions in ticket volume reaching humans within 90 days of deployment.

7. Finance and Accounting: Tropic, Vendr Agents, Anrok, Pilot

Pilot's CFO-as-software service runs bookkeeping, accruals, and management reporting for thousands of startups — outcomes priced, not per-seat. Tropic and Vendr deploy procurement agents that negotiate SaaS renewals autonomously, with Vendr reporting average savings of 12-24% per contract. Anrok automates sales tax compliance across 14,000+ jurisdictions with zero human accountants in the loop for 95% of filings.

The SaaS that loses: per-accountant ERP seats, manual close software, and the AP clerk role. The remaining human role is the controller who reviews agent decisions weekly, not the staff accountant who enters them daily.

8. Customer Success and Health Scoring: Catalyst Agents, Vitally AI, Galileo

Customer success was the last white-collar function to professionalize, and service-as-software is collapsing it back to outcomes faster than any other category. Catalyst's agents monitor product usage, trigger save-plays, run executive business reviews via auto-generated decks, and predict churn at the account level. Pricing flipped from per-CSM seat to per retained dollar of ARR.

The SaaS that loses: CS platforms priced per CSM seat (Gainsight, Totango pricing under pressure), survey tools, and the lower 60% of the CSM job ladder. The remaining human CSM is a strategic relationship owner for top-10% accounts, not a portfolio manager of 80 mid-market clients.

Where Service-as-Software Quietly Fails: Live Collaboration and Decision-Making

There is one cluster where the service-as-software model has not won, will not win in 2026, and may not win in 2030: the synchronous decision-making meeting.

Service-as-software thrives when the output is well-defined (a resolved ticket, a sourced résumé, a drafted contract) and the path from input to output is largely deterministic. Live human collaboration violates both conditions. The output of a strategy session is not a deliverable — it is shared understanding. The path is improvisational. The participants are not optimizing toward a known answer; they are negotiating what the answer should be.

This is why every "AI replaces meetings" pitch since 2023 has died on contact with enterprise buyers. Bots that join calls and transcribe (the Otter/Fireflies generation) are now facing class-action lawsuits and platform-level bot blocks from Google and Microsoft. AI agents that try to participate in live meetings as standalone entities don't have the context to be useful and don't have the consent infrastructure to be trusted.

The category that wins here is the inverse pattern: a synchronous workspace where humans collaborate visually on a shared canvas while AI watches the canvas plus the conversation and adds value in-context. That's Coommit's product thesis — AI as the third teammate inside the meeting, not a service-as-software agent that replaces the meeting.

The takeaway: service-as-software is going to eat the asynchronous, deterministic, single-deliverable parts of your SaaS stack. The synchronous, collaborative, decision-making parts will get more important, not less, as everything around them gets automated.

What Service-as-Software Means for Your 2026 SaaS Stack

If you're running operations at a 50-500 person company in 2026, here are the moves the data supports:

Re-audit your SaaS stack against outcome-priced alternatives. Per Gartner, 30% of SaaS spend is waste. For ticket support, sourcing, legal drafting, and marketing ops, the per-outcome contract is now usually cheaper than the per-seat one. Run the comparison this quarter.

Re-architect for fewer tools, deeper integrations. The service-as-software vendors that win are the ones that absorb 3-5 SaaS categories at once. Your future stack is 30 vendors deep, not 130. We covered the consolidation playbook in detail.

Reinforce the human collaboration layer. As more individual contributor work goes to agents, the bottleneck shifts to alignment, decision-making, and strategy. Invest in the working sessions, canvas-based design, and synchronous review that agents cannot do.

Plan for the pricing flip. Per-seat SaaS budgets are denominated in headcount. Service-as-software budgets are denominated in throughput. Your CFO needs new unit economics — cost per ticket, cost per qualified candidate, cost per closed contract — to evaluate vendors apples-to-apples.

The companies that will struggle through 2027 are the ones still budgeting SaaS as "software for our employees" rather than "outcomes our company purchases." The service-as-software model isn't a vendor-marketing buzzword. It's the way the next generation of enterprise infrastructure is priced — and the way your competitors are quietly cutting cost while you debate whether to buy another seat.