Vidya Raman is a Venture Capitalist. She is a partner at Sorenson Capital where she focuses on investing in deeply technical spaces. Before joining Sorenson she led the AI/ML platform at Cloudera, the fastest-growing product line in the company’s history at the time. There, she was responsible for making ML at scale a reality for customers spanning industries such as autonomous driving, biotech, banking, and government. Before Cloudera, she led engineering and product teams at venture-backed enterprise startups, including eMeter (Sequoia-funded, acquired by Siemens) and Silver Spring Networks (Kleiner-funded, IPO exit).
These days, every VC and founder confidently declares that AI-powered applications are the future, assuming they will dominate enterprise budgets. While this belief is widespread, I see it differently—a perspective not yet popular among VCs and startup founders. While innovation will indeed thrive at higher layers of abstraction, it doesn’t mean a wave of AI startups will monopolize the market. Instead, I believe the real winners over the next decade will be enterprises that invest in DIY applications—custom-built solutions tailored to their unique operations, stakeholders, workflows, and specific business needs.
Why Packaged AI Apps Won’t Be the Only Winners
1. Democratization of Software Isn’t Just for Startups
AI-assisted coding tools have made developing software more accessible than ever, leading to a surge of copycat startups. This wave has increased competition and raised the risk of churn among packaged AI apps. While some startups may initially thrive due to strong distribution channels and network effects, their advantage is unlikely to last.
Importantly, the same tools empowering startups are also available to enterprises—and even system integrators (SIs). Forward-thinking CIOs aiming to cut SaaS spending are encouraging teams to leverage emerging AI-native tools like Langchain, Crew AI, LlamaIndex, AutoGen, among others. This shift signals that all software, whether AI-native or not, will eventually undergo the same scrutiny and face DIY alternatives similar to what SaaS is experiencing today.
2. Enterprise Data is Distributed and Proprietary
In large enterprises, data is often fragmented across multiple clouds, SaaS platforms, and on-premises systems. Much of this data is proprietary and highly sensitive. Adopting new AI applications frequently requires sharing or moving this data, which introduces risks and increases vendor lock-in. Enterprises are becoming increasingly cautious and seeking solutions that allow them to leverage AI without extensive data migration.
The real value of next-generation AI applications will come from enterprises’ ability to fully harness their proprietary data. For example, businesses aiming to utilize their unstructured and structured data must tap into sources like emails, Slack, Zoom calls, internal wikis, Google Drive, Datawarehouses and Lakehouses. The critical question becomes: do you really want 10 separate AI-native apps accessing and managing these sensitive data sources?
3. Powerful AI-Native Developer Platforms Will Emerge
While multi-billion-dollar SaaS companies like Salesforce (founded 1999), Shopify (founded 2006), and ServiceNow (founded 2004) have grown significantly, the rise of PaaS platforms between 2005-2010—such as AWS, GCP, Azure, Heroku (founded 2007), and Vercel (founded 2015)—enabled a new wave of enterprise applications.
At the same time, Developer tools will continue evolving alongside AI adoption. The next generation of AI-driven developer tools will empower enterprises to build custom AI applications instead of relying solely on packaged solutions. Emerging AI-native tools like Poolside, Tessl, Vercel’s V0, Stackblitz, and Cursor offer enterprises new opportunities to create tailored AI-powered experiences.
4. Enterprises Will Build Some of the Most Impactful AI-Native Applications
With advanced AI models like Deepseek, Llama, and Mistral deployable on local PCs or private servers, enterprises now have fewer reasons to depend on third-party vendors for AI infrastructure. This opens the door to a new generation of on-premises applications—solutions traditionally ignored by startups.
Deploying AI applications directly on-device or at the edge will benefit industries such as healthcare, finance, and manufacturing by enhancing security, reducing latency, and allowing greater control over sensitive data. For example, hospitals can use AI for real-time diagnostics without exposing patient data to the cloud, while enterprises can streamline workflows without transferring data across multiple cloud environments. Additionally, orchestrating enterprise data, fine-tuned models, and agents is better suited for internal development than third-party packaged apps.
5. Shiny New Apps Aren’t Always Necessary
Enterprises have accumulated vast amounts of legacy code and countless customizations across various software products over the years. Most modern startups avoid this complexity—and for good reason. However, a significant cost-saving opportunity lies in using AI to analyze, understand, and automate the maintenance of these existing workflows and applications.
AI-powered tools like Moderne and Mechanical Orchard can help enterprises optimize legacy systems, reduce technical debt, and enhance operational efficiency. This transformation will primarily be driven by in-house IT teams aiming to modernize internal systems without adding new vendor dependencies.
Risks and How to Mitigate Them
1. Enterprises Stuck in the Pilot Phase
Many enterprises fall into “pilotitis”—an endless cycle of experimenting with AI without deploying it at scale. While ongoing experimentation has been necessary due to rapid changes in AI tooling, it’s crucial to set time limits on pilots and deploy small but meaningful use cases in production to learn and evolve effectively with AI.
2. Unclear TCO Will Not Last Forever
Enterprises often find piloting AI solutions easier than scaling and maintaining them in production. While it may seem that packaged apps are better suited for this challenge, history tells a different story. In previous technology waves, early adopters like Yahoo and Google led the charge in big data, ML, and AI before mainstream enterprise adoption followed. Similarly, companies like Netflix spearheaded microservices and containerization before they became industry standards. Enterprises must approach AI adoption cautiously while maintaining a sharp focus on total cost of ownership (TCO).
Conclusion
While AI-native packaged applications will gain traction, they won’t be the only drivers of AI innovation. The shift toward local AI deployments, the rise of powerful developer tools, and enterprises’ need for flexibility will make internal AI development just as vital. The future of AI isn’t limited to apps—it lies in enabling enterprises to build, deploy, and scale AI on their own terms.