I love this debate - it's been brewing in my head for years.
For the last 15+ years, the SaaS world has basically worshipped at the altar of "Opinionated Software." You know the gospel: Build something with strong convictions, force the user to do things your way because you, the brilliant founder, know best. Constraints breed creativity. Say "no" to feature requests. It was preached by DHH at 37signals, embodied in Basecamp's ruthless simplicity, echoed by Apple’s walled garden, and it powered the early SaaS explosion. Tools like Linear, Superhuman, and even Notion (in its calmer days) felt like a breath of fresh air after the bloated enterprise monsters of the 90s and 2000s.
And, it worked. It really did. We got polished, fast, delightful products that didn't drown us in options. But lately... I'm exhausted just thinking about it. In 2025, I genuinely believe this religion is cracking. The "opinionated" era is hitting a very real wall, and AI is the hammer smashing it.
The core promise of opinionated software was that one mental model could rule them all. But reality hits hard once teams grow beyond 10-20 people, or when businesses aren't cookie-cutter startups anymore.
Real-world examples pile up everywhere:
What happens? We don't change the software - we warp our companies to fit it. We build Zapier skyscrapers (those fragile towers of automations that break if anyone breathes on them). We hire "ops wizards" whose full-time job is translating real work into what the tool allows. I've felt this pain personally in past companies: months lost forcing square-peg processes into round-hole tools.
As companies scale, the diversity of processes explodes. Opinionated software assumes homogeneity; reality delivers chaos.
I've been on both sides of this fence. As a founder/builder, opinionated software feels like survival. Your roadmap is already a war zone: endless feature requests, angry support tickets, competitors cloning your thing overnight (hello, new Notion clones every week). Customers beg for "just this one tiny thing," but you know adding it starts the slide toward bloat. Say yes too often and you become a Bloatware - 1,000 features, nobody is happy.
Competition is brutal now. Open-source replicas, AI wrappers, geographical clones - they pop up faster than you can ship. Keeping a tight ship with strong opinions is how you stay focused, move fast, and build something people rave about. Disgruntled customers? They hurt. Every "this doesn't fit us" email feels personal, as if you'd failed them.
But here's the raw truth from the builder trench: pure opinionated design is becoming a luxury we can't afford forever. The next winners won't just slap a customer-support chatbot on top that gives canned answers or routes tickets. They'll go deeper with native AI, not just as a feature but a foundation that delivers true per-customer personalization while guarding a rock-solid core.
That's the magic AI unlocks. Not gimmicks. Real, instance-level malleability.
AI changes everything because custom engineering just got dirt-cheap.
Six months ago, giving every $49/mo customer their own dev was insane. Today? AI agents can refactor schemas, rewrite UI, adjust logic - all at near-zero marginal cost.
We're entering Malleable SaaS:
Picture this (and yes, people are already prototyping it):
You say: "We run 6-week cycles, not quarters—remake the OKR dashboard."
The tool doesn't hide fields or add a hacky custom field. It refactors: new DB tables, updated reports, redesigned views. Safely, versioned, rollback-able.
This isn't sci-fi. Look at emerging research and builds:
We finally get both worlds: the polish and speed of SaaS + the perfect fit of bespoke software.
I'm pouring my soul into Avery (avery.dev), an AI Virtual Engineer that makes this possible. To demonstrate, we are also building a malleable business software suite - Avery Software (avery.software) - a unique software platform where every customer lives on their own "branch" - isolated, safe, infinitely tweakable by AI. No more shared-roadmap begging. Need a feature? Ask your instance. It builds, tests, and deploys. The "anti-roadmap."
It's scary - security nightmares, code safety, keeping the core coherent - but the upside feels massive. If we nail this, we free builders from roadmap hell and users from workaround purgatory.
I'm not alone in feeling this shift. The walls between no-code, low-code, and AI-coding are crumbling. If you're an indie hacker or founder wrestling with this:
I disagree with the "AI makes opinionated software more important" crowd (sorry Contrary Research) - AI makes rigidity a liability. The tools that bend with users will eat the static ones.
If this resonates, hit me up in the comments or DM. Let's build the post-opinionated future together. I'm exhausted by rigid tools, excited as hell about what's coming next.
Building in public at avery.software and avery.dev
Opinions win, but that's just my opinion.
Simplicity brings certainty. Crafting simplicity requires an understanding of complexity, so it can be avoided. AI tools incur uncertainty because they are complex products.
Good design is what you leave out.
I completely agree – opinionated software can create simplicity.
The nuance for me is that simplicity depends on context. For consumer apps, “simple” can often be defined in a fairly universal way.
For business apps, though, every company works differently, so simplicity starts to mean fit and adaptability. A SaaS product can feel simple to its creators, yet be painful for a specific business if they have to twist their workflow to match the software’s opinions.
Opinionated UX works brilliantly when you are serving a very tight niche where everyone operates in roughly the same way. Outside of that, the ability to adapt becomes part of what “simple” really means.
This is one of the clearest articulations I’ve seen of why “opinionated SaaS” is running into structural limits. What you describe as the “opinion gap” is exactly what I’ve watched play out in multiple industries — not because founders are wrong, but because real-world workflows mutate far faster than any rigid tool can.
What really resonated is the idea that companies bend themselves around software, instead of the other way around. I’ve seen entire operations teams whose full-time job is essentially translating messy human processes into the narrow tunnel a tool allows. Zapier skyscrapers is the perfect phrase.
Where I’m still wrestling is this: how do we preserve the reliability of opinionated systems while allowing AI to mutate logic safely?
Versioning is one answer, but schema drift, permissions, integrations, and data integrity all explode in complexity when each tenant becomes its own branch.
I don’t doubt the direction — the user-adapted instance is coming — but I’m curious how you think about the governance layer. If AI can reshape models and UI on demand, what prevents entropy over time? Or do you see a future where “controlled entropy” becomes a feature, not a problem?
Either way, the shift you’re describing feels inevitable. Rigid tools worked when businesses were more homogeneous; today variability is the norm. The teams that figure out safe malleability will define the next decade of SaaS.
Thanks for the thoughtful comment. You nailed it: the biggest challenge of malleable software is absolutely maintaining reliability amidst that flexibility.
To answer your question on entropy: We do believe the future is indeed 'controlled entropy.' We are approaching this by separating the architecture into 'core-locked' components (reliability) and 'AI-malleable' layers (flexibility). The safety net comes from what we call code-retrace and code-reconciliation - essentially allowing the AI to mutate logic while maintaining a strictly versioned, rollback-capable lineage.
It’s a massive research area in itself, but essential. Will share more updates as we progress with our R&D.
Thanks for the detailed breakdown — really appreciate you taking the time to explain your thinking.
The separation between core-locked components and AI-malleable layers makes a lot of sense, especially combined with versioned rollback. That “controlled entropy” approach seems like the only viable way to keep flexibility without losing reliability.
Totally agree that the reconciliation layer is where a lot of the future challenges (and opportunities) will sit.
Curious — have you found any practical limits yet in how far the AI layer can mutate logic before hitting stability issues?
The key to AI stability is absolutely tight context engineering. This is an evolving space in its own right.
This is a smart take — AI is pushing SaaS toward flexibility, meaning software may finally adapt to users instead of forcing users to adapt.
Malleable SaaS could reshape the software landscape by giving users more flexibility, but it also challenges the value of “opinionated” software that guides workflows and best practices.
Why not both? Opinionated on core features and extensible/customisable on client specific features, i.e. 3rd party integrations but make it AI.
This is already happening today, 3rd-party integrations via the likes of Zapier. But that's not the same as the software itself getting personalized to the individual business workflows. The core features are the most important aspect of the business workflow, and having that opinionated makes it a one-size-fits-all approach. I am not saying that this doesn't work - just that it is not optimal, as it adds a learning curve for a business on top of other things.
Good point. This article really challenges the dogma of “rigid, opinionated” SaaS — arguing that thanks to AI, software will increasingly adapt to the user, not force users to adapt to it. If that plays out, we may be entering a much more flexible, personalized era of SaaS.
Thought-provoking! Malleable SaaS offers flexibility, but it challenges the idea of opinionated software—balancing customization with guidance will be key for users and developers alike.
Great point — that balance between freedom and guidance is exactly where things get interesting. Too much flexibility can overwhelm users, but too much opinionation limits what they can build. The challenge is finding that middle layer where the system stays coherent while still letting people shape it to their needs.
Curious — have you seen any products that strike that balance well?