The hype cycle is settling. Here's an honest read on what's working, what's been oversold, and where the real opportunities are sitting in the intersection of AI, SaaS, and software development.
Two years into the current AI wave, the landscape is starting to stabilise in ways that are useful to talk about honestly. The teams that chased every AI announcement have a lot of half-deployed experiments. The teams that waited for clarity have real adoption gaps to close. And the teams that picked a few high-value use cases and executed well are pulling ahead in ways that are hard to catch up with. Here's what we're actually seeing across the AI, SaaS, and software development work we do with clients.
AI Is No Longer a Feature — It's Baseline Infrastructure
Twelve months ago, an AI chatbot on your SaaS product was a differentiator. Now it's expected. Customers who use HubSpot, Notion, Linear, or any major SaaS product in 2026 have been trained to expect AI-assisted suggestions, automated summaries, and intelligent search as table stakes. The bar has moved. For SaaS founders, the question is no longer "should we add AI" — it's "which AI capabilities will we ship in the next two quarters and how do we do it without introducing unreliable behaviour that damages trust."
How Dev Teams Are Using AI in Their Own Workflows
- Code generation and completion — engineers report saving 1–2 hours daily on boilerplate, test writing, and documentation with tools like GitHub Copilot and Cursor
- PR review assistance — AI-assisted code review catches more surface-level issues, freeing human reviewers for architectural and logic-level feedback
- Spec-to-ticket generation — product managers are using LLMs to turn PRD documents into structured Jira tickets in minutes
- Test generation — AI-generated unit and integration tests have meaningfully improved test coverage in teams that were previously under-tested
- Incident runbooks — AI-generated runbooks from past postmortems are becoming standard in DevOps workflows
The productivity gains from AI in software development are real and consistent — we see them across every team we work with. The risk is over-reliance: engineers who lean on AI for code generation without deeply reviewing output are introducing subtle bugs that are harder to catch than the kind they'd write themselves.
What's Overhyped vs. What's Genuinely Useful
Overhyped
- Fully autonomous AI software engineers replacing development teams — current tools assist, they don't replace; complex reasoning and system-level understanding still require experienced engineers
- No-code AI app builders producing production-ready software — great for prototypes and internal tools, not for scalable products with complex data models and security requirements
- "AI will run your entire business" automation frameworks — these age poorly when business logic changes and the AI has no way to know what changed or why
Genuinely Useful
- AI-assisted customer support with human escalation paths — the best implementations we've seen deflect 65–75% of tickets and handle the rest faster
- Document and data extraction pipelines — high accuracy, high volume, well-defined inputs and outputs; ideal for AI
- AI-enhanced search within products — semantic search dramatically outperforms keyword search for knowledge bases, documentation, and content libraries
- Automated reporting and anomaly detection — AI that tells you something changed and why is saving analytics teams significant time
- Developer tooling integration — the productivity gains are well-documented and compounding with each generation of tools
SaaS Product Decisions That Matter More Now
- 1Data quality before AI features — AI features that work on bad data are worse than no AI features; fix the underlying data model first
- 2Explainability for trust — users need to understand why the AI made a suggestion, especially for consequential actions; add context and reasoning to AI outputs
- 3Opt-out paths — some users will not want AI-assisted anything; build opt-outs early or you'll be building them under pressure later
- 4Cost monitoring — LLM API costs are unpredictable at scale; build token usage monitoring and per-feature cost tracking from day one
- 5Eval pipelines — you need to know when a model update or prompt change degrades your feature; build automated evaluation before your users find the regression
The SaaS companies winning with AI right now are the ones treating it like any other critical infrastructure: instrumented, monitored, tested, and understood by the team maintaining it — not treated as a black box they've plugged in and hoped for the best.
Where the Real Opportunities Are
The most interesting space in 2026 is the intersection of AI and vertical SaaS — industry-specific applications where there's rich structured data, domain expertise requirements, and established workflows that AI can genuinely improve rather than just approximate. Healthcare, legal, construction, logistics — these verticals have been slow to modernise and the combination of AI and thoughtful product design has more room to create real value here than in horizontal tools that are already commoditising.
