Why Now?
The SaaS model made sense for two decades. It solved real problems: no servers to manage, automatic updates, accessible from anywhere, pay-as-you-go pricing. Companies gladly traded control for convenience.
But the math has changed. Three converging forces have made 2026 the inflection point for SaaS repatriation:
- AI has made your data exponentially more valuable, and more dangerous to give away
- Self-hosted systems are more legible and operable for agents than SaaS product surfaces
- AI agents will take the system operations burden that justified SaaS in the first place
The Data Problem
Every SaaS vendor is now an AI company. They have to be. It's existential. And AI companies need data. Your data.
Read the updated terms of service. Look for phrases like "to improve our services" and "aggregate insights." Your documents, your conversations, your workflows. They're all training data now.
This isn't necessarily malicious. Vendors genuinely believe they're making their products better. But the result is the same: your competitive intelligence, your proprietary processes, your customer data. It's all feeding models that serve your competitors too.
The irony is sharp. You're paying for the privilege of giving away what might be your most valuable asset.
Vibe Coding Custom Solutions
This is where a lot of the attention is right now. And for good reason.
The traditional software procurement chain is comically lossy. A customer has a problem. They explain it to a sales person. The sales person translates it for a product manager. The PM writes a spec for an engineer. By the time code gets written, the original signal is buried under layers of interpretation, politics, and roadmap prioritization.
Now imagine short-circuiting that entire chain. The person with the problem sits down with a coding agent and builds exactly what they need. Not a generalized solution for a market of thousands. A purpose-built tool for their specific use case.
But this isn't the complete picture. DIY works great for CRMs, internal dashboards, workflow tools. It doesn't work for your database. Your observability stack. Your identity provider. Some problems are genuinely hard and benefit from dedicated teams working on them full-time.
The repatriate AI movement includes: vibe-coded custom solutions where they make sense, and self-hosted open source or commercial software where they don't.
Agentic System Operations
The initial argument for SaaS was operational: "We don't have the headcount to run this ourselves." That argument is evaporating.
AI agents in 2026 can:
- Monitor infrastructure and respond to incidents
- Apply security patches and manage updates
- Scale resources based on demand
- Debug issues with full context awareness
- Migrate data between systems
- Configure and maintain applications
They're not replacing engineers. They're starting to handle the operational toil that made self-hosting prohibitive, and it will only get better.
We expect to see an open repo of skills emerging that help agents get even better at operating these applications & the underlying infra. Today this is the territory of side projects and homelabs but as models progress and skills emerge, this will eventually manage critical infra.
The Migration Path
What actually locks you into SaaS isn't the software itself. It's how fully configured & integrated everything is. Years of dashboards, alerts, SSO setups, integrations between systems. The thought of recreating all that is what keeps people paying.
Just like operations, we need specific agent skills for each migration. Not generic "move my data" instructions, but targeted playbooks: migrate-datadog.md, migrate-okta.md, migrate-salesforce.md.
Each skill needs to understand what to extract:
- Configuration: Dashboards, alert rules, SAML/SSO settings, user roles, workflows
- Data: Historical records, logs, documents, customer information
- Integrations: Webhooks, API connections, sync rules between systems
Some migrations are config-only. Moving from Datadog to Grafana is mostly about recreating dashboards and alert rules. The metrics data itself often isn't worth migrating.
Others are data-heavy. Leaving Salesforce means extracting every contact, deal, activity, and custom field. Plus all the automation rules built up over years.
The hardest part is often the integrations between systems. Your CRM talks to your email, which talks to your support desk, which talks to your billing. These connections need to be mapped, understood, and rebuilt.
Agents Thrive with Root
The case for repatriation is not just cost, sovereignty, or privacy. It is that repatriated software is more legible and operable for agents.
SaaS usually exposes a product surface: the safe, narrow layer built for human admins, multi-tenant controls, and abuse prevention. Repatriated or self-hosted systems expose a system surface: logs, databases, queues, config, infra, and often source.
Agents do better on systems than products because they can inspect state, trace failures, bootstrap users, and recover from broken flows without waiting on vendor-sanctioned paths.
Even basic onboarding reveals the difference. SaaS products often can't allow arbitrary account creation because trust establishment is part of the product: signup flows, abuse controls, email verification, billing, identity, rate limits. In self-hosted systems, trust can be established by the operator or environment. If you have root, you can create the first admin, seed the tenant, reset auth, or patch the bootstrap path entirely. What is a product constraint in SaaS becomes an operational decision in self-hosted software.
The infrastructure form factor doesn't matter. A VPC on AWS. A bare metal server at Hetzner. A rack in a colo. A Mac Mini in your office. What matters is: do you have root? If you don't have root, you're renting access to your own data, and your agents are locked out of the control plane.
The Gaps
Let's be realistic. Not every piece of software needs to be self-hosted today. Some categories still make sense as managed services.
CDNs. Payment rails. SMS gateways. Some compliance tools. These are areas where the infrastructure is genuinely global, the regulatory burden is high, or the economies of scale are real. We still see value here.
The agent skills to configure, migrate and operate still don't exist. We see the need for a common spec & validated library of these skills. This could be part of the vendor's secret sauce, but selling markdown files is going to be tricky. Most likely these will be open source with community contributions.
Open source is going to change a lot. The current maintainers are at odds with the new model. It isn't clear where things will go but slop forks, spam PRs, hallucinated fixes are already causing problems. We expect projects to evolve with architectures that allow users to run upstream for updates and consistent agentic management, while extending the functionality they need through vibe-coded plugins and modules.
The Role of Software Vendors
The industry is going to undergo massive creative destruction. It won't be overnight, but the underlying financial model of valuing software companies as a multiple of sales due to their previously ever expanding and renewing revenue base, is coming to an end. However, with some imagination you can see a world where leaner software vendors still thrive while both producing and capturing value. These vendors will adjust early and rethink their business models. Some areas of enduring value that we expect to see:
- Network Effects: Clearly if a vendor is able to maintain a clear network effect from transactions, trust, data etc they will be able to retain that in the new AI era.
- Services: Vendors are often subject matter experts, consultants with tools that they can apply to specific business problems and processes. This knowledge will be increasingly valuable the question of what to build is more important than how to build it.
- Responsibility: Vendors have long been a means to shift responsibility of a problem externally. "One throat to choke" is an old saying that will remain true, just being willing to take the blame when things go wrong is worth a lot in larger orgs. The assignment of liability, indemnification, compliance assurance etc are all economically valuable and will come at a cost.
- SLAs & Support: Even with AI agents operating the applications, vendors will be able to provide guarantees for uptime, response times and general escalation. Assurance and insurance on critical operations has always been valuable and remain so.
- Benchmarking to SOTA: If vendors are able to produce clear benchmarks against SOTA models and harnesses, they should be able to capture some of the value of that improvement. This will be measured on multiple vectors including effectiveness, speed, cost etc, and will need to be constantly updated as the SOTA evolves.
What's Next
This is just the beginning. The self-hosting renaissance will accelerate as:
- Agents get better: Operations will become nearly autonomous
- Local AI matures: Running frontier-quality models on commodity hardware
- Open source closes gaps: The few remaining weak spots will fill in
- Regulations tighten: Data sovereignty requirements will force the issue
- Talent shifts: Engineers want to work with systems they can own
The SaaS era gave us convenience at the cost of control. The AI era gives us the tools to have both.
Kill your SaaS. Repatriate your stack. Own your future.