In which we look at current generative AI trends and discussions, and try to figure out what’s going on.
The Year of the Agent
2025 was the year of the Agent, at least in gen AI nation. It seemed like all of a sudden, everything was all about agents and agentic AI. Everything, everywhere, all at once – about agents, what they can do, what they can’t do, and how they are revolutionizing computing.
One big debate is about whether AI agents can be truly autonomous, or whether a human in the loop (HIL) is always going to be required.
This gets to the heart of what an AI agent is – meaning that it automates an LLM chat sequence. An agent is given a role, a task, some instructions or constraints, and a context within which to work.
Agents can reduce an LLM’s rate of hallucinations and incorrect results by constraining the scope of activity and simplifying an iterative validation of the responses to prompts, and updating prompts to improve responses.
We see an explosion of interest in agents across any and all types of software systems. Existing enterprise products develop and ship AI agent capabilities, positioning their existing technologies as best suited to meet demanding enterprise requirements for security, reliability, scalability, performance, and reliability.
A cadre of new, agentic-only software systems sprouting out of university research projects and startups reimagine agents as independent entities, collaborating primarily with each other to complete their tasks instead of adding AI to existing capabilities.
Some vendors embrace the generative AI sea change – emphasizing agents for applications better suited for non-deterministic outcomes – while others reject the idea of overly non-deterministic outcomes and define guardrails and policies to eliminate or reduce non-deterministic behavior as much as possible.
And of course a swarm of agents quickly gathers on demand whenever needed to generate and validate new and modernized applications.
Meanwhile in Seattle and London
Or somewhere, anywhere really. Take Seattle for example.
A restaurant owner vibe codes an entire restaurant application in five days and saves on software product fees forever. Isn’t this the real story? (Granted, as he says, he has a background as a software engineer, which is not typical of a restaurant owner.)
But then you have an AI services company in London that encourages all of its staff to use AI coding tools to create new applications and release them to the public.
Not just the developers, but the product managers, sales managers, marketing managers, administration staff, etc. It’s apparently that easy if you are trained (which they are).
So if anyone and everyone can create entire applications in a day or a week using gen AI coding tools, why is anyone buying any software at all anymore?
The SaaS-Pocalypse, SaaS-aPalooza, or SaaS-aParilla
Or something that makes a sound like the SaaS product market is exploding, imploding, popping, or otherwise sinking.
Forbes published an article about this, starting with this sentence:
“SaaS company valuations in the first week of February 2026 saw a massive sell-off. In seven days, over $1 trillion in market capitalization was erased from software stocks.”
So the market is apparently already starting to factor in the impact of gen AI coding tools on SaaS company valuations as individuals create their own applications and reduce or eliminate SaaS licensing and subscription fees.
But how realistic is it to conclude that gen AI coding tools will sink the SaaS product market?
Or is this just more of the same gen AI hype we’ve been getting since ChatGPT descended the escalator in November, 2022?
Does Historical Context Help?
In the earliest days of enterprise computing, business applications were created individually for a single business.
Andit was very expensive. Only certain businesses could justify the investment, which was typicaly calculated by comparing the cost of a manual process to an automated process, and determining how much you could save by having a computer automate the manual process.
Kind of like the ROI calculations we are now doing to figure out how much it costs to manually develop an application vs how much it costs to automatically code generate the application.
Anyway, at some point software vendors realized they could use the same application for multiple businesses, save money by reusing something they had already developed, and pass on a potion of the savings to their customers, which expanded the market.
And eventually of course the PC came along and software companies started to develop personal applications for them. These had an even bigger profit curve (anyone remember the 80% profit after covering your R&D costs equation?) due to the much larger potential customer base.
Eventually PC applications became like books or CDs, basically just publishing items for consumers, manufacturing thousands if not millions of copies and generating billions of dollars in profits for “hit” titles.
But enterprise applications are not so easy to standardize. Companies compete on the basis of their differentiated operating models, and this means their core business applications often have unique requirements.
For example, the Amazon’s retail site’s fast shipping application became a competitive advantage over other retail sites that could take a week or more to ship an order. Other retail websites are catching up, or have caught up, but Amazon had to create this application themselves – you could not buy a software system like that on the open market at the time.
Similarly in the stock trading area, investment banks had to develop their own applications to be competitive in high frequency electronic trading, squeezing milliseconds out of the trade processing roundtrip to gain a price advantage often worth millions. You also cannot buy such an application.
And as a final example, consider the Tesla application that plans your trip, finding the nearest free charging station along your route, and steers you to it. This is a complicated process involving real time telemetry collection from the vehicles and the charging stations, calculating the best routes, and communicating back in real time to the drivers.
Fitting your Business to the App (or Vice Versa)
On the other side of this application coin are the non-strategic business processes – things a business can do in a number of ways without impacting their competitive position in the market.
It doesn’t impact competition very much how you do payroll, or manage vacation time, send invoices, pay bills, or keep the general ledger up to date. (At least not for most businesses – I’m sure there are some exceptions where even these applications can have a competitive aspect. But anyway.)
For applications like these lots of businesses can use the same one out of the box. It makes sense for software companies to invest in productizing them, since it’s a volume play and margins can be significant once costs of development, packaging, sales and marketing etc are covered. And it makes sense for businesses to buy them because it would be more expensive to build them and there would be no competitive advantage in doing so. Also the software vendors amortize the development and maintenance costs across multiple customers, putting the cost for an individual company well below what it would cost to develop an equivalent application.
When a company buys a standard software product for a standard business function, the company often may have to adjust its internal processes to conform to the way the software product works, but again this usually is cheaper than building the applicaiton from scratch. Or at least it used to be.
During the early days of ERP system availability (especially SAP) this was a common complaint. ERP systems allowed some level of customization, since business processes tend to vary, but customization often wasn’t easy and could be expensive. Many organizations discovered that it was cheaper and easier to change their business process than customize the ERP system. But now maybe they don’t have to even perform this calculation.
The Intellyx Take
What if instead of buying software, a business could just custom develop everything? Will the examples from Seattle and London hold true generally? Or just for small business and personal applications?
When generating your own applications you may not have to even think about customizing standard software or adapting your business process to an ERP or other software package (such as a CRM and other SaaS package) in order to use it.
This is what leads to all the dire predictions these days about how SaaS is dead, now that anyone can create any application they want using generative AI coding tools.
On the other hand, though, there’s a growing body of evidence indicating gen AI coding tools are not sufficient for large, enterprise scale applications. If your application is larger than one person can prompt for or manage, vibe coding is not likely to work.
If your application requires enterprise architecture oversight, or high levels of scalability, performance, reliability, or security, it’s unlikely a program manager will be able to generate it successfully.
In any case, gen AI coding tools (including agents) are going to be very helpful in meeting an almost endless list of requirements for new standard applications, but it will not be simple to use them successfully for creating or modernizing mission critical and strategic applications.
As usual with gen AI, the answer here appears to be that “your mileage may vary” and “it depends” very much on what you are doing. Generative AI is not a panacea, despite how wonderful it is, it’s better at some things than others. And it remains to be seen whether it can truly replace SaaS, or any other complex, mission critical business application.
Copyright © Intellyx BV. Intellyx is an industry analysis and change agent advisory firm, covering enterprise IT from mainframes to artificial intelligence. Our broad focus across technologies allows business executives and IT professionals to connect the dots among disruptive trends. No AI was used to write this article. Image credit: Google Gemini.
