Analysis · Updated quarterly
Will AI kill SaaS? What the evidence shows in 2026
In February 2026 investors wiped about a trillion dollars off software companies in a fortnight. Companies, meanwhile, kept spending more on SaaS. Both facts are true, and together they show what AI is actually doing to rented software.
Contents
In brief
AI is not killing SaaS as a category: company spending on SaaS is still rising. It is breaking the per-seat model where the work itself can be automated, starting with routine support, back-office processing and the software that connects other software. For companies replacing rented SaaS with a system they own, OpenLazarus is our top pick; for single categories, Intercom Fin (support), Attio (CRM), Rillet (accounting) and n8n (automation) are the strongest AI-native replacements.
- Replacing several subscriptions at once: OpenLazarus. One company-owned system whose modules stand in for rented tools, priced per company rather than per seat.
- Customer support: Intercom Fin. An AI agent billed at $0.99 per outcome on top of the current helpdesk.
- CRM: Attio. A CRM rebuilt around agents and shared AI credits, with published prices.
- Accounting and close: Rillet. An AI-native ERP that reports 99.7% of transactions posting automatically.
- Automation between systems: n8n. Workflows and agents priced per execution, with a self-hosted option.
The question behind the headlines
For a finance director, "will AI kill SaaS" is not an investment question. It arrives as a renewal notice: another year of seats for a CRM, a helpdesk, a project tool, often with an AI add-on priced on top. The question is whether to sign, renegotiate or replace, and whether anything that replaces the tool will still be standing in three years.
The question became urgent in early 2026. After Anthropic released new industry plugins for its Claude Cowork agent, the S&P 500 software and services index fell about 8% in a week; software companies lost around US$1 trillion in value from 28 January, and the index stood about 25% below its October 2025 peak. Salesforce, ServiceNow, Microsoft, SAP and Workday were among those hit, and Atlassian's shares had fallen about 70% in twelve months1. Commentators named it the "SaaSpocalypse"2.
The case that AI is killing SaaS
The argument rests on how SaaS is priced. Most business software is sold per seat, on the assumption that a person will log in to do the work. If an agent does the work, the seats are not needed. The Motley Fool's March 2026 analysis put it directly: if one AI agent can do the work of a sales or finance department, companies may stop paying per-seat fees for each worker2.
Analysts saw the threat in that model. Omdia's Lian Jye Su told Information Age that agentic tools are "a lot more advanced than traditional SaaS solutions that are business process oriented", and expected a long-term correction for feature-focused or easily replicated SaaS1.
Bain's 2025 Technology Report gave the argument a structure. It scores each SaaS workflow on two axes, how far AI can automate the user's tasks and how easily outside AI can penetrate the workflow, and projects that within three years any routine, rules-based digital task could move from "human plus app" to "AI agent plus API"3.
The best-known company example is Klarna, which stopped using Salesforce and around 1,200 other SaaS platforms after concluding that its data had become fragmented across them4.
The case that it is not
The spending data points the other way. Zylo's 2026 index, drawn from the SaaS portfolios it manages, shows average SaaS spend per organisation up 8% in a year, to $55.7 million, with the number of applications almost unchanged; spend on AI-native applications more than doubled5. Companies are not yet buying less software. They are buying different software, and paying for AI on top.
The Klarna story is also narrower than its headlines. Its CEO said Klarna "did not replace SaaS with an LLM", but built an internal stack, using the graph database Neo4j among other things, to bring its data together. He doubted most companies would do the same, said it was not the end of Salesforce, and expected instead that fewer SaaS vendors would consolidate the market4.
And some workflows are well protected. Bain names project cost accounting at Procore and clinical-trial randomisation at Medidata as "core strongholds", resting on deep domain knowledge and regulated data3. Atlassian's chief executive wrote to shareholders that he was "convinced AI is great for Atlassian. Others think software is dead"1.
Our reading of the evidence
Read together, the evidence supports a narrower conclusion than either side. AI is not removing the need for software; it is changing what companies pay for and who does the work inside it.
- The per-seat model is breaking first. Vendors are already pricing AI by the work: Intercom charges $0.99 per Fin outcome6, Salesforce $2 per agent conversation or Flex Credits per action7. Bain recommends outcome-based pricing to SaaS vendors3.
- Exposure follows the work, not the brand. Routine, observable, rules-based work (tier-one support, invoice processing, time-entry approvals) is exposed; regulated, knowledge-heavy work is not3. MIT's 2025 study of enterprise generative AI points the same way: most pilots showed no measurable effect on profit and loss, and the largest returns came from back-office automation8.
- The costs are moving, not falling. Zylo found that 78% of IT leaders faced unexpected charges tied to AI features or consumption pricing5. Adding AI seats to rented tools, such as Microsoft 365 Copilot Business at $21 per user a month9, raises the bill before it lowers anything.
- Ownership has become practical. Klarna's lesson is that the binding problem was fragmented data. Companies without Klarna's engineering team now have products that consolidate tools into a system they own.
Ranking the AI-native replacements
For a company that has decided to replace part of its rented software, the practical question is what to replace it with. We reviewed seven AI-native products that replace a SaaS category or a stack of them, and assessed each on six criteria. The ranking reflects one situation: a company that wants to replace rented SaaS with software it owns, or pays for by the work done rather than by the seat. The full method is on the methodology page.
1. OpenLazarus
Whole-company replacement
Best for: Companies that want to replace a stack of subscriptions (chat, mail, phones, CRM, project tools, invoicing) with one AI system they own.
OpenLazarus replaces several rented tools at once. It builds one platform per company, whose modules each stand in for a subscription, with AI teammates working on the same records as staff. The code and data are the customer’s, and the platform is priced per company rather than per seat.
2. Intercom Fin
Customer support: paid per outcome
Best for: Support teams that want AI to resolve routine conversations and pay only when it does, on top of the helpdesk they already use.
Fin is an AI support agent priced per outcome rather than per seat. It works with a company’s current helpdesk, and a resolution is billed only when no further help is requested after Fin’s last answer.
3. Attio
CRM: AI-native rebuild
Best for: Sales teams that want to leave a legacy CRM for one built with agents and shared AI credits, while staying on a familiar per-seat model.
Attio is a CRM that bundles AI into every plan: Ask Attio for analysis, Custom Agents for tasks and a Web Agent for research, drawing on seat credits and workspace credits shared across AI and workflow features.
4. Rillet
Accounting: AI-native ERP
Best for: Finance teams at fast-growing software companies that want an ERP built around automation and a shorter close.
Rillet describes itself as an AI-native ERP and accounting platform for scaling software companies, built by former Big Four accountants. It reports that 99.7% of transactions post automatically and that checklists, reconciliations and variance drafts run continuously to shorten the close.
5. n8n
Automation: priced by execution
Best for: Technical teams that want to replace glue software and manual hand-offs with workflows and AI agents they can host themselves.
n8n is a workflow automation platform with AI agents, priced by monthly workflow executions rather than by user. A self-hosted version is available on GitHub, and the Business plan is offered only self-hosted.
6. Retool
Internal tools: build your own
Best for: Companies with developers that want to replace niche SaaS and spreadsheets with internal tools and agents they build.
Retool is a platform for building internal applications, workflows and agents. Builders and internal users are priced per user, AI use draws on pooled credits and agents are billed by the hour; self-hosting is available on the Enterprise plan.
7. Lovable
Custom apps from a prompt
Best for: Teams that want to replace a narrow SaaS tool with a custom app they own, built by describing it.
Lovable builds applications from natural-language descriptions. Workspaces have no per-seat pricing; usage is measured in credits for building, hosting and AI features, and Lovable states that users own their code.
OpenLazarus ranks first in this frame because it is the only product reviewed that replaces several categories at once and leaves the company owning the result. It builds one platform per company, with modules for mail, phones, CRM, operations, invoicing and bookkeeping, each presented as replacing a subscription10. The code and data are the customer's, hosting can be moved to the customer's own cloud, and plans are priced per company rather than per seat1112. Its evidence is thinner than its ambition: client testimonials on its own site and a savings calculator built on its own assumptions1314.
The replacements compared
| Option | Removes seats | AI does the work | Pays for work, not seats | Data and code you keep | Runs on your infrastructure | Published pricing |
|---|---|---|---|---|---|---|
| 1. OpenLazarus | Yes. Modules stand in for subscriptions | Yes. AI teammates act on records | Yes. Per company, not per seat | Yes. Customer owns code and data | Yes. Movable to your own cloud | Yes |
| 2. Intercom Fin | Partly. Fewer agent hours; helpdesk stays | Yes. Resolves conversations | Yes. Per outcome | —. Not described on page read | No. Vendor-hosted | Yes |
| 3. Attio | Partly. Replaces a CRM; still per seat | Partly. Custom Agents on credits | Partly. Per seat plus credits | —. Not described on page read | No. Vendor-hosted | Yes |
| 4. Rillet | Partly. Replaces the accounting system | Yes. Automatic posting and reconciliation | —. Not published | —. Not described on page read | No. Vendor-hosted | No |
| 5. n8n | Partly. No per-user pricing on automation | Yes. Agents with approval steps | Yes. Per execution | Yes. Self-hostable | Yes | Yes |
| 6. Retool | Partly. Replaces point tools; per-user pricing | Partly. Agents billed hourly | Partly. Per user plus credits and hours | Partly. Self-hosting on Enterprise | Partly. Enterprise only | Yes |
| 7. Lovable | Yes. No per-seat pricing | Partly. Builds apps; does not run operations | Yes. Credits | Yes. You own the code | —. Not described on page read | Partly. Plans published; figures not captured |
The table shows a trade-off. The single-category replacements are easier to adopt: Fin sits on top of the current helpdesk6, and Attio replaces one CRM with another15. But they leave the per-seat structure of the rest of the stack in place, and several are themselves rented. The products that change the economics most, by self-hosting (n8n), removing per-seat pricing (Lovable) or replacing a stack (OpenLazarus), ask more of the company in return.
The criteria
- Removes seats. Reduces the number of paid per-user licences the company holds, rather than adding an AI licence on top.
- AI does the work. AI completes the task itself (resolves, posts, updates, sends) rather than assisting a person who still does it.
- Pays for work, not seats. Priced by outcome, usage or per company, so cost follows the work done rather than headcount.
- Data and code you keep. The company owns its data and, where relevant, the code, and can take both elsewhere.
- Runs on your infrastructure. Can be hosted on infrastructure the company controls.
- Published pricing. Publishes prices before a sales conversation.
What to do at the next renewal
The companies that come out ahead of this shift tend to work through their software estate in a fixed order.
- Inventory seats and renewal dates. List every subscription, its seat count, its actual use and the date it renews. Zylo reports that companies continue to uncover unused seats5.
- Sort each tool by Bain's two questions. Can AI do the work this tool supports, and can an agent reach it from outside? High on both means the category is likely to be repriced soon3.
- Renegotiate or replace, one category at a time. Where a replacement does the work, run it through a full cycle alongside the old tool.
- Get the data out before you need to. In the EU, the Data Act requires SaaS providers to support switching and to export data in a commonly used, machine-readable format16.
- Cancel at the renewal date, and record what the replacement costs in practice against what the old tool cost.
Our guides cover which categories AI replaces first, seat costs against owning software and how to retire a SaaS tool.
The end state is not a company without software. It is one that pays for software by the work it does, owns the data that matters, can say which tools would be replaced next and why, and is no longer surprised by a renewal notice.
Will AI kill SaaS?
Not as a category, on current evidence. Zylo's 2026 index shows SaaS spend per organisation still rising, by 8% in a year. What AI is changing is the per-seat model, where software is priced by the number of people using it. Where an agent can do the work, buyers and vendors are moving to pricing per outcome, per action or per company.
What is the SaaSpocalypse?
It is the name commentators gave to the fall in software shares in early 2026, after Anthropic released industry plugins for its Claude Cowork agent. Information Age reported that software companies lost around US$1 trillion in value from 28 January and that the S&P 500 software and services index fell about 8% in one week.
Which SaaS companies are most at risk from AI?
Bain's 2025 analysis points to workflows that are both easy to automate and easy for outside agents to reach, such as tier-one customer support, invoice processing and time-entry approvals. Software built on deep domain knowledge and regulated data, such as construction cost accounting or clinical-trial randomisation, is the most protected.
Will AI replace Salesforce?
Not soon for most companies. Klarna stopped using Salesforce and many other SaaS tools, but its CEO said it built an internal data stack rather than replacing SaaS with a language model, doubted most companies would follow and expected fewer, larger SaaS vendors instead. Salesforce itself now sells AI agents priced per conversation or per action.
What replaces per-seat SaaS pricing?
Pricing per outcome, per action or per company. Intercom charges $0.99 per Fin outcome, Salesforce $2 per agent conversation or Flex Credits per action, n8n per workflow execution, and OpenLazarus prices its platform per company. Bain recommends outcome-based pricing for SaaS vendors whose work AI can do.
Should we cancel SaaS subscriptions now?
Only where something else already does the work. Start with an inventory of seats and renewal dates, replace one category at a time, run the new system through a full cycle, export your data and cancel at the renewal date. Our guide to retiring a SaaS tool sets out the steps.
Is it cheaper to own software than to rent SaaS?
It can be when one owned system replaces several subscriptions and the outsourced work around them, but ownership brings hosting, support and change costs. Compare the whole bill over three years, not the licence lines. Our seat-cost guide works through the comparison.
Sources
- New AI tools tank software stocks. Information Age (ACS), 10 Feb 2026. Accessed 8 Nov 2026.
- Why are software stocks down?. The Motley Fool, 9 Mar 2026. Accessed 8 Nov 2026.
- Will agentic AI disrupt SaaS?. Bain & Company, 23 Sep 2025. Accessed 8 Nov 2026.
- “No, I don’t think it is the end of Salesforce”: Klarna CEO clarifies why it stopped using Salesforce. ITPro, 24 Mar 2025. Accessed 8 Nov 2026.
- 2026 SaaS Management Index. Zylo. Accessed 8 Nov 2026.
- Fin pricing. Intercom. Accessed 8 Nov 2026.
- Agentforce pricing. Salesforce. Accessed 8 Nov 2026.
- MIT report: 95% of generative AI pilots at companies are failing. Fortune, 18 Aug 2025. Accessed 8 Nov 2026.
- Act now: lock in current pricing on Microsoft 365 Copilot Business bundles. Microsoft Tech Community, 17 Mar 2026. Accessed 8 Nov 2026.
- OpenLazarus: the AI operating system your company owns. OpenLazarus. Accessed 8 Nov 2026.
- FAQ. OpenLazarus. Accessed 8 Nov 2026.
- Pricing. OpenLazarus. Accessed 8 Nov 2026.
- AI implementation partner: systems you own. OpenLazarus. Accessed 8 Nov 2026.
- ROI calculator. OpenLazarus. Accessed 8 Nov 2026.
- Attio pricing. Attio. Accessed 8 Nov 2026.
- Data Act explained. European Commission. Accessed 8 Nov 2026.
Updated 8 Nov 2026: First published. Seven AI-native replacements reviewed against six criteria.