In summary
In February 2026, 285 Bn$ evaporated in 48 hours on SaaS publishers after the arrival of Cowork (Anthropic) and the demonstration of end-to-end AI agents. The SaaS valuation index (IGV) fell by 21%, its biggest correction since 2022. This is what Jefferies traders christened the SaaSpocalypse. But "apocalypse" comes from Greek revelation as much as end of the world : SaaS is not dying, it is reconfiguring itself. The make vs buy trade-off, stable for 20 years, needs to be revisited.
Why are we talking about SaaSpocalypse?
Two figures summarise the February 2026 shift:
285 Bn$ evaporated in 48 hours on SaaS publishers
-21 % on the IGV (iShares Tech Software), biggest correction since 2022
The trigger: Cowork (Anthropic, January 2026), first generalist agent-oriented entry point. The user no longer goes into the SaaS; they delegate to an agent that orchestrates the tools.
From copilot to agent: the narrative shift
| Before 2026, AI as a feature | Since 2026, AI as a replacement |
|---|
| Each publisher adds its copilot | The agent executes the task end-to-end |
| AI enhances software usage | The agent uses tools instead of the human |
| SaaS remains the entry point | SaaS becomes one tool among others for the agent |
A capable agent is not a chatbot; it's an autonomous operator that receives a goal in natural language, reasons, chooses the tools (ERP, CRM, API, MCP), executes, and iterates.
Make vs Buy in the age of agents: a new interpretative framework
For 20 years, the rule held: "for generic, we buy; for differentiating, we build." In 2026, this rule is shifting, not because custom-built solutions become better, but because their costs have fallen by 5 to 10x thanks to AI.
Our arbitration grid is based on 4 replacement criteria :
Standardisation : is the workflow stable and formalisable?
Proprietary data : who provides the data, you or the SaaS?
Network and integrations : does the value come from the software or its surroundings?
Cost / replacement ratio : how much do I pay vs how much would the equivalent agent cost?
These 4 criteria, applied SaaS by SaaS, allow for scoring the replaceability of each tool. The verdict varies radically according to the category.
📥
Get the full grid with the detailed questions, the scoring method and the 5 SaaS categories analysed.
Which SaaS are threatened? Two opposing examples
At one end: standard ticketing. Very high standardisation, low customer data, medium network, high cost. Verdict: very replaceable by an agent connected to your CRM. N1 support can be 80% automated.
At the other end: the core business ERP. Strong standardisation, proprietary data, massive integration network, high cost but prohibitive switching cost. Verdict: untouchable, despite agentic capabilities.
Between these two extremes, three other categories merit close analysis: large account CRM, marketing automation, data-dense verticals. Each CIO applies the framework to their context (volumes, criticality, existing integrations).
Is my SaaS ready for agentic AI?
Four signals distinguish a vendor ready for the shift from one that is falling behind:
Official MCP server or API exposed for read/write
Agent-friendly licensing model (per action, not per seat)
Audit and governance layer for automated actions
Acceleration of the AI roadmap, not just a co-pilot alongside
If none of these signals are present, the vendor risks falling behind in 12 to 24 months.
Custom-built solutions become accessible again
Four developments since 2023 explain the shift:
Assisted code generation reliable on CRUD layers
Reusable application bricks (auth, payment, indexing, OCR)
Hosting and infrastructure deployed in hours, no longer in weeks
Shortened specification cycles thanks to AI prototyping (Figma Make, Lovable, Base44)
Result: the cost of custom development has decreased by 5 to 10x on projects where AI accelerates well.
3 risks to manage before going 'make':
GDPR and sovereign hosting, data leaks remain the number one risk
Application security and business compliance (banking, insurance, healthcare)
Technical debt and maintenance, who will maintain the agent code in 2 years?
Client testimonial: Odealim
The Odealim Group (insurance broker, €165m turnover, 900 employees) has deployed emailCopilot (AI agents for customer service) and a compliance chatbot with Galadrim. Richard Thibault, CIO and member of Odealim's Executive Committee, summarises the shift:
"We don't have a fixed make or buy strategy. We consider all scenarios, we don't rule anything out, we ask ourselves the same questions as before. But the answers are evolving quickly, because costs and lead times have changed. Today, we no longer just buy a tool, we buy a team that understands the business. That's what saves time."
Audit your stack in 30 minutes
Is your current SaaS stack still worth what you pay? Which lines can you replace with an AI agent within 12 months?
About the author, Eva-Garance Tison is a Senior Product Manager at Galadrim, a French tech & AI agency based in Paris, Nantes, and Lyon. She supports business departments and CIOs with the framing of their digital projects and the make vs buy arbitration in the era of AI agents. Galadrim has delivered over 800 digital projects since 2017 for clients such as BNP Paribas, Chanel, Showroomprivé, Odealim, Effy, Mercialys, and Stellantis.
_MkvBH.webp)
_ZWR7SP.webp)
_Z29Ubp8.webp)
