Introduction
The SaaSpocalypse is indeed real. The rise of AI has brought a new competitor – that of a custom vibe-coded solution. Those without the right product are at significant risk of disruption. Hundreds of millions of people now use LLMs on a daily basis around the world and user taste has materially changed – especially with the cost to build and customize using LLMs is a fraction of what it once was. This post outlines what “good” looks like for product in an AI-first era with evolved LLM-centric user taste.
For those skeptical of vibe-coded solutions, here’s a proof point – I have a colleague running a well-established UX/product building agency (Pixelspace). They have literally replaced almost all commercial software with in-house built solutions. Although a vibe-coding implosion will probably happen (likely over security) someday – the threat is indeed a clear and present danger for the next few years at minimum.
Bad UX will No Longer be Tolerated
The last 15+ years have seen many B2B SaaS solutions ship with truly sub-par UX, ranging from awkward to atrocious. And in many cases mobile was an awkward afterthought or missing entirely. SaaS providers were able to get away with this because they were often the only game in town or the competition was simply no better.
With the ability to build and integrate virtually effortless compared to what it used to be, people will simply build their own replacements with the UX they desire – often-times on top of an incumbent SaaS platform’s APIs, instantly defeating the per-seat pricing model that has underpinned SaaS for several decades.
The lesson here – if the UX is in need of love, give it the love it deserves. It’s time for a refresh or a user revolt cannibalizing per seat pricing becomes an imminent risk.
Fragmented Portfolios are also on the Outs
Many SaaS companies have grown through M&A. Almost as many have never done solid integration leaving many products with highly disparate user experiences. Users have to jump around from product-to-product or module-to-module to get things done and live with a highly fragmented experience.
Likewise, this also simply isn’t going to be tolerated by users any longer – it’s a recipe to start vibe coding integrated UXs and workflows. Don’t make someone jump around and complete a workflow in 60 clicks when they can do it in 5 or ask an agent via chat to do it for them.
APIs Must Be First Class
The best cannibalization is to oneself. Apple wrote the book on this and few companies have been as successful at great execution. Nobody cried in Cupertino as iPod sales dwindled to zero in lieu of iPhones and iPads.
For a SaaS platform to survive and thrive, the best offense against vibe coding is to simply be the best platform building block available for a given problem domain. To do this, world-class APIs covering the ENTIRETY of a platform are required. API surfaces must be modern; REST is so yesterday…
GraphQL should be the default choice for developers building in a “pull” scenario. You can get all you need in a single query enabling high efficiency on a per call basis. Webhooks are needed for the “push” scenario. Data can be pushed to subscribed endpoints and manipulated as needed. The combination represents the modern “power duo” for which anything can be built easily and from any platform.
World Class MCP must Overlay the APIs
MCP, or Model Context Protocol, is the agreed upon industry standard for exposing platform capabilities to agents such as ChatGPT, Claude, or Gemini. Exposing the ENTIRETY of a platform via MCP is critical to preventing disintermediation in a vibe coding world. The prerequisite of MCP is world-class APIs.
When implementing MCP, the tendency is to do this in a quick-and-dirty fashion, which kind of defeats the purpose and will just end up frustrating users. Instead, build it out entirely ensuring a full repertoire of Tools, Prompts, and Resources covering the breadth of the platform AND chat-based use cases. This will require thinking through how users will interact with the platform via chat end ensuring these cases are well covered as part of the MCP implementation.
Reports and Dashboards are Out; Natural Languate Queries are In
Almost every SaaS platform offers some type of analytics interface. This has been a mix of pre-exising dashboards and report creation and customization tools. Although this will need to continue to exist, the new preferred means of achieving business intelligence is natural language queries.
Just as one can crunch amazing amounts of data with any of the frontier LLMs and conduct massive deep research with the likes of Gemini’s Deep Research or Claude’s Research modes – users now expect the same thing in their SaaS applications.
The danger though is simply connecting a LLM to raw data can result in egregious conclusions and dangerously confident hallucinations. This creates an opportunity for SaaS vendors to offer their own solutions that can, for example, use existing data pipelines and queries to prevent inaccuracies and hallucinations while still offering the natural language query users now desire.
My Portfolio is Guilty of One or More of These Points; what do I do?
The good news is that the cost of building and evolving product with AI (provided that the right AI SDLC practices are in place) is lower than it ever has been in the history of software development. Tackling a massive portfolio makeover is no longer the proverbial task of climbing Mt. Everest that it was even 18 months ago.
Having just been through such a journey and with the gained knowledge and battle scars, here are some real-world recommendations on how to tackle it:
- Start with APIs – this is the underlying technical dependency that literally will unlock everything else discussed herein (or make it significantly easier). If I could rewrite history on my own journey, I would have done this first.
- MCP Comes Second – With great APIs in place, shipping a world-class MCP implementation is the best combination defensive and offensive move one can make. Done well enough, one can expose their platform in the MCP marketplaces in ChatGPT, Claude, and Copilot. This takes the pressure off because now people frustrated with your platform will be building their replacements on top of your platform instead of going it solo or down the street to a competitor.
- Natural Language Queries is Third – With the MCP wall in place, this is the next area to focus on as it is the area likliest for things to go seriously wrong with LLMs and again have users start rethinking their platform choices. And, it requires great APIs to be in place.
- Fixing UX and Portfolio Fragmentation Comes Last – This is going to be the biggest, messiest, and most expensive (albeit exponentially cheaper to affect now than even a short while ago). It’s a “must-do” for holding onto any seat-based licensing model, provided that 100% cannibalization in lieu of APIs and MCPs isn’t an acceptable business outcome. Exploring and deciding this tradeoff is someone that can only be done within one’s walls (although hopefully backed by solid customer research) But given the price tag, it’s best left for the end after the defensive and offensive moves have been played first and the proverbial moat secured. (For what it’s worth, in my last journey we started here; real world lesson would sequence this last.)
What’s next?
In subsequent posts, I’ll explore the commercial opportunities the SaaSpocalypse is now creating; the financial shift that has occurred affecting anyone with investor expectations around value creation; and all of the people and process dynamics required to refactor one’s product and technology organizations for this new realm. More to come…
Leave a Reply