The Application Layer Is Melting: Why Your Software Stack Is Becoming Invisible

Enterprise software's biggest shift since the cloud isn't about new features it's about making interfaces optional.

I've been tracking enterprise software trends for years, but something fundamentally different happened in 2025, it is unprecedented and it is happening so fast. It wasn't just another acquisition announcement, it was a signal that the entire application layer “as we know it” is about to melt away.

Here's what I mean.

The $8 Billion Wake-Up Call

Salesforce just dropped $8 billion on Informatica, and honestly, the price tag isn't even the crazy part. What caught my attention was Marc Benioff's explanation: they're building "a unified architecture for agentic AI" so autonomous agents can "operate safely, responsibly, and at scale."

Reading between the lines? Salesforce doesn't think you'll be clicking around software much longer. They're betting on a future where AI agents make decisions for you, and those agents need bulletproof data to work with.

This isn't just another enterprise acquisition it's Salesforce admitting the UI is becoming obsolete.

And they're not alone.

Your Apps Are Quietly Becoming APIs

The signs have been everywhere if you've been paying attention. HubSpot just launched the first CRM that plugs directly into ChatGPT. Now their customers can skip the dashboard dance and just ask "show me my best converting leads from last month and build them a nurture sequence."

Box did the same thing they built a secure bridge so ChatGPT can rifle through your company's documents, contracts, and strategic plans. Instead of hunting through folders, you can ask the AI to "analyze our Q3 performance against industry benchmarks using our internal reports."

See the pattern? The UI is becoming optional. The API is becoming everything. Your favorite SaaS tools are quietly transforming into data sources for someone else's AI. Check out my earlier article about MCPs

The Tool Explosion Problem

Meanwhile, Scott Brinker's annual martech map just hit 15,384 tools up 9% from last year despite economic headwinds. We added 2,489 new tools while 1,211 others got acquired or simply died.

This creates a brutal economic reality: why sign three-year contracts for Tool #9,037 when an AI can query your data warehouse on demand? The old "buy vs. build" calculation just got a third option: "ask an AI to figure it out."

CFOs are starting to ask the uncomfortable question: What's the ROI on software that requires humans to operate it?

Why "Chat With Your Data" Falls Short

Everyone's rushing to build "chat with your data or documents" features, but the excitement fades fast when you hit real-world problems:

Context chaos: When the AI mentions "ACME Corp," which one? The subsidiary, the parent company, or that customer record from 2019? Without proper entity resolution or Identity management between internal & external data, you're basically rolling dice.

Trust issues: Can you really bet quarterly decisions on AI insights when you can't see the math? Business-critical stuff needs explainable AI that shows its work.

Surface-level understanding: Most systems can find relevant documents or search is undoubtedly easy but still misses the deeper business logic or root cause. Does your AI know that churn risk spikes after NPS drops below 6, or that customers who contact support exactly twice are 40% more likely to upgrade?

The difference between a useful AI system and an expensive hallucination machine comes down to one thing: knowledge engineering.

The Unsexy Truth About AI Success

The companies getting AI right aren't just building cooler chatbots. They're doing the unglamorous work of organizing their knowledge properly.

Think about business knowledge built on structured & unstructured data in three buckets:

Explicit Knowledge: The stuff you're already tracking CRM data, spreadsheets, inventory counts or its available via third party data or external networks. If it's in a database, it's explicit.

Tacit Knowledge: The tribal wisdom that never makes it into documentation. Like how Sarah from sales always knows which prospects are serious, or why customers who ask about pricing upfront rarely convert.

Emerging Intelligence: Hidden patterns that only surface when AI analyzes massive datasets. Maybe there's a connection between geographic location, purchase timing, relationships in your prospects accounts and churn that no human would think to look for.

Salesforce's massive Informatica bet essentially buys them a shortcut to the first two layers, positioning them to unlock the third at scale. Although i am not sure how two legacy companies can create a modern agile tech and culture. But if you look at it they're not just buying software they're buying a structured understanding of how business data connects & its lineage.

The Three-Phase Transformation

Smart companies are approaching this systematically:

Phase 1: AI Assistants
Deploy ChatGPT, Claude, or Copilot alongside existing workflows, tools & MCPs. Focus on individual productivity and getting people comfortable with AI. Popular GPTs or current Agents are good templates to use.

Phase 2: Specialist AI Apps
Replace specific expert tasks with purpose-built AI. GitHub Copilot for code, specialized tools for Marketing Ops, data quality, AI video creation. These integrate into existing workflows but handle high-value tasks autonomously.

Phase 3: Agent Networks
Deploy AI systems that own entire business processes pricing optimization, customer intelligence, retention campaigns, supplier negotiations. This is where you get compounding gains, learning loops and real margin improvements.

Here's the catch: none of this works safely without solid data foundations. Skip the boring work of data modeling and ontology design, and you'll spend all your time fighting AI hallucinations.

The Hard Questions Leadership Won't Ask

As this accelerates, most leadership teams are asking the wrong questions. Instead of "Where can we use AI?" they should be asking:

What knowledge gaps are blocking AI from delivering transformative insights? The biggest wins often come from solving foundational data problems, not adding more AI features.

Does each vendor in our stack offer clean APIs that autonomous agents can use? If a tool only works through human interfaces, it's becoming a liability.

How do we flip our training budget? Stop teaching people to navigate complex interfaces or Adobe, Salesforce, 6Sense experts. Invest in data modeling instead. Your competitive advantage isn't how fast humans can click it's how well AI agents understand your business.

What happens when AI agents start making purchasing decisions? B2B sales might soon be AI negotiating with AI, starting with discovery & search while Buyers may have their agents too. Are you ready for that world?

The End of Clicking (If You Want It)

Enterprise software's next decade belongs to companies that turn databases into decision-makers via what we call in iCustomer via decision Intelligence. The traditional application layer buttons, menus, dashboards isn't disappearing, but it's becoming secondary to natural language conversations with intelligent agents.

As HubSpot's Dharmesh Shah put it when announcing their ChatGPT integration, it's "a match made in heaven" between conversational AI and “unified customer data”.

Companies that recognize this shift early and invest in the foundational infrastructure clean data, robust ontologies, agent-ready architectures while leveraging their existing first party data foundation via composable way will have an insurmountable advantage.

The transformation is already happening. I'm seeing it in every enterprise software roadmap, every vendor pitch, every IT strategy meeting. The question isn't whether your software stack will become invisible to users.

It's whether you'll lead the transition or scramble to catch up.

---

What are you seeing in your organization? Are you already planning for an agent-first world, or still fighting to get basic integrations working? Hit reply or DM me and let me know. I read every response.