The Future of SaaS is a Hybrid of Agent & Platform + Vertical AI Innovation
AI is not killing enterprise software, but changing it
Hi all, I’m trying to learn more about how AI is being adopted in the creative industry. Please DM me if you have good insights, read good reports, or know the right people to talk to. (This is coming out this week as well, stay tuned.)
Today’s post is a sequel to China’s startups pivot to vertical AI piece and follows our thought process on why China and the U.S. chose divergent AI monetization strategies - hint* China in consumer application end, and the U.S. focused on enterprise software.
Microsoft CEO Satya Nadella famously claimed in a BG2 podcast that traditional business apps (SaaS) will soon be replaced by AI agents. This statement sent the internet into a frenzy, with headlines like “Satya said SaaS is dead.” But as the argument has played out over the last few months, we’re realizing that that’s not exactly what he meant.
Basically, he claims that traditional SaaS apps, which are essentially CRUD (Create, Read, Update, Delete) databases with business logic, are becoming obsolete because AI agents will take over the business logic layer.
He envisions a future where AI agents manage workflows and decision-making across multiple applications and databases, effectively replacing the traditional SaaS model with what he calls "Agents-as-a-Service" (AaaS). This shift means SaaS apps will revert to simple data stores, while AI layers provide the intelligence and orchestration.
In a subsequent interview with Indian entrepreneur Varun Mayya, Nadella said that in his world, Excel and Word are going to be agents that go with the co-pilot, a completely new way of thinking about the workflow. And he’s not the only one thinking this way. Greg Isenberg, CEO of LateCheckout, has also been very vocal on this front, and we will explore his points below.
And despite macro challenges and still a very unclear ROIC to AI applications right now, VCs and startups are still charging the way and they appear undeterred. In fact, they seem more convinced than ever about AI’s long-term role in software. Leading VC firm Andreessen Horowitz just announced plans to raise a $20 billion megafund focused on AI, doubling down on AI investments, especially in application use cases, a shift away from just infrastructure investments.
As I wrote in my piece on China’s AI startups, a similar trend is taking off in the U.S. We’re seeing a decisive shift: startups are moving away from competing on frontier models and doubling down on vertical use cases. And I think there actually is more open disucssion about this in the U.S. right now also because of the different AI monetzation strategies the two countries have chosen - China in consumer application and U.S. betting on enterprise adoption.
Historically, SaaS vendors have had a tough time selling to SMBs. These businesses are far more cost-conscious, every tool needs to deliver either concrete cost savings or clear revenue upside. As Primary VC put it, “SMBs are tough to sell to and even harder to retain.” However, enterprise software is probably even harder to replace.
So where is the narrative of “AI will replace SaaS” getting its confidence?
AI Transforming SaaS
The Shopify CEO message leak went viral last week and reignited a discussion that’s been simmering for months: Are AI tools replacing new hires? A while back, the prevailing narrative was that SMEs were freezing headcounts and relying more on GPT-like tools, while large enterprises were still hiring and integrating AI gradually. However, now even Shopify appears to be signaling a shift.
This has sent some frights and shivers across the industry, but another argument that has come up is that if we’re not bringing in new talent, are AI tools actually good enough to replace their capabilities? Will AI agents be able to replace human talent, or to extent, can they replace certain workflows that previously required humans to complete? What is the future of AI in SaaS usage and adoption? Is AI killing SaaS?
Satya Nadella said that in the future, when hiring talents, “on a prosaic level,” he means that people will have “a basket” of agents that they’ve built, and they’ll be hired not just for their competence but also for their agents. He actually acknowledges that SaaS won’t be replaced completely but will be complementary to how we do things now.
On the other side of the argument. Greg Isenberg, CEO of LateCheckout, took to LinkedIn last month and has been a vocal leader in the argument that SaaS is dead. And he shared his view that AI agents will dismantle SaaS in three phases.
What we’re mostly seeing now can be considered as AI as Co-Pilot, mostly in a supporting role to our existing SaaS operations.
In the next year or so, we will see AI as Autonomous Operator which it will be able to bypasses human interfaces, unbundles workflows.
And then in 2-3 years time, we’ll be in an era of “software invisibility” where AI agents will talk directly to APIs and bypass the human interfaces altogether.
He predicts that “this democratization of software creation means every company becomes a potential software producer rather than just a consumer.” Companies will eventually abandon subscription-based tools in favor of custom-built, AI-generated software. Ultimately, leading to SaaS (as we know it), will then just all die.
However, this vision oversimplifies how technology is actually adopted in the enterprise. It overlooks practical realities like slow adoption - incrementalism, the importance of data analysis and systems of record, and the sometimes inefficiencies and unsophistication of DIY software development.
Instead, what I think will happen is that SaaS will be forced to evolve into a hybrid model where vendor platforms and custom-built AI agents will coexist, much like what Satya Nadella has said.
Potential Trends in Enterprise Adoption
1. Data Analysis is Not Easy
Many existing platforms, such as CRM service providers, accounting platforms, and HR software, are actually systems of record where they store, track, structure, and help analyze enterprise data, whether that is customer information or sales figures.
Even if agents can help receive instructions and execute tasks, these mega-service platforms will remain critical in an AI-native world because enterprise data cannot simply exist in thin air or be disorganized in random folders on our desktops.
And beyond the storage, analyzing the data is a capability that a stand-alone AI agent will not be able to do. This is where Satya’s point about the agent will coexist with the existing softwarethat we know of, like, Excel or Word, makes more sense. Proprietary data and the systems of record will continue to serve as foundational layers for the company’s processes and operations, just with AI intelligence added on top.
2. Vibecoding + DIY
The phenomonon of vibecoding has allowed many non-technical players to build their own apps and build on top of existing available products. But frankly, these DIY projects probably aren't good enough for mass-scale adoption if it’s not done properly by a professional.
The thought of being able to build your own apps is great, and many have shown that you can do it. But at an enterprise level, it also ignores the operational burden of building, maintaining software and scaling. Enterprise adoption of tools will need to be adopted across firms to ensure processes alignment, and it requires upkeep and trouble-shooting. To be honest, it may work well for SMEs but not for large corporations when you need to control quality across markets and time zones.
Added on, DIY may make collaborations or cross-market projects more challenging, and they are definitely not things companies want to mess with. A AI-native application is different in this case.
(On top of this, you’re assuming people are still willing to pay for the base product first - which we’ve written it’ll be a massive hurdle in China given the open-source options there are now. Might work better in the U.S.)
3. Hybrid Baby: SaaS + AI
The most likely outcome is that existing SaaS and AI will give birth to some kind of hybrid. SaaS companies have already started embedding GenAI into their products (think Canva, Figma), opening up APIs for agents and serving as secure foundations, and this will only continue. Customers will then build custom logic on top for edge cases or proprietary processes.
In Bernstein’s recent report, Software: GenAI application winners/losers framework, backs this up as well. Highly customizable software that mirrors enterprise workflows becomes staff augmentation, which creates deep stickiness. In contrast, general-use AI tools often require enterprises to retrain their processes to fit the software, which creates friction. Switching costs in SaaS aren’t just about money; they’re often about process change. And that favors existing, trusted vendors with strong integration capabilities.
And to Nadella’s point, Microsoft will thrive in this new world, because businesses will likely still buy standard tools and build on top of them. Standardizing that within their operations to make it most business-friendly for their case and, in many ways, this is not that different from many of the SaaS offerings out there right now.
This hybrid model offers the best of both worlds: vendor reliability and customization flexibility.
4. AI Adoption Will Not Be Overnight
Lastly, any enterprise adoption will not be overnight, unlike the consumer phenomenon we’ve seen. Over the last two years since OpenAI’s ChatGPT release, consumer adoption and for the sake of this argument - vertical adoption has been rapid. Still, enterprise AI adoption isn’t going to be gradual and layered due to bureaucratic reasons and system processes, data security, talent and uptraining time, and switching costs.
LLMs excel at tasks focused on ideation and summarization but still often falter on deterministic tasks like accounting or forecasting. And let’s not forget the hallucination risks. Although we’ve written about how AI value creation will shift to application-end, that was more about consumer-facing applications.
Most enterprise adoption will more likely happen in three phases:
AI assists users within existing tools (e.g., summarizing CRM activity, media monitoring, streamline information). This basically has already been implemented across consulting firms;
AI takes on more tasks in an agentic way, but still under close supervision to ensure logical reasoning and accuracy, we’re seeing this with agents like OpenAI’s Operator or ManusAI;
And then it’s the supposed “AGI”, AI autonomy, but that likely will still be within structured, known workflows that have been proven by human labor - at least for the near-term.
[And I wrote about what tasks I think AI agents can replace and why certain tedious tasks should be done by AI in PR]
This approach reflects organizational realities because enterprises often have the mentality if - “dont fix what is not broken.” So, in many ways, that sets up enterprise incumbents to be well-positioned to embed AI incrementally and preserve (at least try to retain) customers and help them with AI upgrades in existing workflows. It will be hard for new players to challenge their existing reach and deep industry know-how, and even user behavior.
The Other Race: Vertical SaaS
Then, there is the discussion on vertical AI. Vertical AI startups are gaining traction for good reason. And we’ve talked about how in the China AI startup space, most of not all have pivoted from LLMs to a vertical use case amid DeepSeek’s breakthrough.
In the same interview with Varun Mayya, Satya Nadella said the competition for companies will extend beyond LLMs and be in their ability to make breakthroughs in building AI agents for various industries.
Competing on foundational models is expensive and distribution-intensive. Instead, founders are chasing faster ROI in high-value, industry-specific use cases. Instead of a mass-market SaaS integrating AI, this is building a tool for a niche market or at least a very specific industry use-case, such as the financial services sector, law, healthcare, or even specific tasks within an industry, like the social intelligence tool Meltwater for PR and communications professionals.
Jason Shuman, General Partner at Venture Capital firm Primary, who specializes in AI vertical investment, wrote on LinkedIn that founders are working on the same vertical SaaS use cases for AI at “breakneck pace”.
The three most important things right now are 1) Speed, 2) distribution advantages, and 3) strategic product decisions. (which we’ve explored in details in consumer-facing apps) And in many ways, the same logic applies for SaaS.
Especially where speed, distribution leverage, and smart product sequencing matter most. It’s about building moats fast, testing quickly, and thinking two steps ahead.
Trace Cohen at a family office echoes this: “The future is vertical AI—deeply specialized startups solving high-value problems in industries like healthcare, finance, and defense.” These players combine domain expertise, proprietary data, and utility-driven design. Even if the market size is smaller, the adoption curve is steeper. This is echoed in the Chinese AI market.
Conclusion: AI won't Kill SaaS
This brings me back to Greg Isenberg’s post from above, which suggests that AI will replace all our existing SaaS tools. I think he has some fair points, but that vision underestimates enterprise adoption challenges and is definitely over-optimistic about corporate efficiency and openness to change. I am more aligned with what Satya Nadella is saying.
AI probably won't replace software in the near term, but it will rewire the anatomy of these platforms and workers' behaviors in using these tools. As RMIT Professor Jonathon Boymal recently pointed out as well, research shows that Microsoft 365 Copilot is changing work habits of knowledge workders but it’s in areas like “email management” that is “adjusted independnetly” and not in “behaviors requiring coordination, like meetings,” again to my point - scaled adjustments in the processes across enterprises will take time but independent efficiency gain or productivity enhancing tasks will adopt AI at a much faster scale.
On AI Proem, we’ve explored many consumer AI applications and their speedy adoption rate and why, but for enterprise SaaS, the future isn’t “AI vs. SaaS.” I think it’ll be SaaS that behaves like AI and AI that becomes infrastructure. And the future belongs to those who (human or company) adapt and hence also my repeated emphasis on the importance of how we educate our next generation about AI.
SaaS vendors that open their platforms to AI agents, retain control of data pipelines, ensure customer security, and empower customer innovation will thrive. What we’ll see is more startups focusing on vertical use cases or creating agents on top of existing SaaS to complement the existing workflow. What that means is we might have more players than we originally imagined and incumbents like the big techs would likely thrive in consumer end or use cases of large distrubtion and reach. Meanwhile, newcomers will have their chance to lean on their industry expertise.
Thank you for your attention and interest, please hit subscribe or share with a friend if you found this useful/ insightful/ entertaining. Also feel free to leave a message or DM me if you think I’m spewing absolute gibberish. Cheers!