Schema, Shema: Context for the Win

The 30-year playbook of the application software industry is dead.

Lovable AI's meteoric rise and Airtable's recent AI product launches aren't just cool features—they're the canaries in the coal mine signaling the complete destruction of how we've thought about application software since the 1990s. What's happening beneath the surface is far more profound than another AI wrapper: the fundamental assumptions about fixed schemas and implementation ecosystems are crumbling in real-time.

Context now trumps schema in almost every way. The traditional dynamics of buying, deploying, and extracting value from application software are forever changed.

Does Metadata Even Matter Anymore?

A year ago, I was convinced metadata was the moat. More metadata meant more context for your agentic assistant. More context meant better AI. Better AI meant a defensible advantage.

That thesis is now becoming dead on arrival.

The emergence of Model Context Protocol (MCP) and other LLM-native integration services means any system of action can easily access your precious metadata from any system of record. Your carefully curated data structures and relationships? They're now commoditized APIs that any competitor can tap into—if you let them.

The new delineation is crystal clear: Systems of Records may have interesting metadata, but Systems of Actions actually get shit done. They grab whatever context and data they need from wherever it lives to deliver the jobs customers hired them to do.

Your metadata isn't your moat—your ability to act on it is.

The Breakneck Speed of AI's Evolution

We've witnessed the fastest transformation in software interfaces in IT history:

Phase 1: AI Helps Us Use Software (The Co-Pilot Phase) Instead of hunting and pecking across multiple screens and tabs, we ask AI to assist us, answer questions, and complete tasks. This felt revolutionary 18 months ago.

Phase 2: AI Does the Work (The Agentic Phase)
AI agents now perform entire roles with minimal guidance—roles we previously hired full-time employees to execute. This transition happened in months, not years.

Phase 3: AI Creates the Software (The Vibe Coding Phase) This is where it gets truly insane. I just spoke with a founder of ours who spent 1/10th the time creating a new product module than his team had spent debating whether they should pursue that direction. He and his CTO built it over a weekend without bothering their 75+ strong engineering organization. They plan to launch it next week—a product they believe will command a 50% upcharge to their existing solution.

One weekend. 50% upcharge. 75 engineers bypassed entirely.

The Death of the "We Bring Schema, You Hire Armies" Industrial Complex

For time immemorial, a software company set out to serve a given set of jobs for a company. It had a database schema that reflected the underlying CRUD requirements for those jobs-to-be-done, along with the associated views and workflows. Application software companies were horizontal (think SFA) or vertical (think HVAC-specific order and project management).

For a horizontal offering, it was bought by a company (which itself is in a vertical business) that then hired consultants and admins to customize the product to ensure it met their vertical and company-specific needs—what I call their "context." Over time, this required too much configuration and expense for many industries, so we saw the emergence of vertical or industry-specific editions from those horizontal software companies, making them much easier to market, sell, and, in theory, deploy and derive value from.

Over the last two decades, we have seen the emergence of vertical-specific software companies that offer multiple pieces of software tailored to their target customers' specific needs, allowing those customers to rely on the vendor for all their software requirements. Even then, consultants and administrators were required, perhaps fewer than for horizontal deployment, but still in large numbers.

Regardless of the flavor or version of software being marketed, this is somewhat ass-backwards if we're honest with ourselves.

Companies want to run their businesses better and run them their way—what I call their "context." They shouldn't have to spend $250 billion annually on professional services to implement software and another $500 billion annually on operations and administrative personnel to make it consume their context on an ongoing basis. Yes - $750B spent across the application software industry annually, just so companies can insert their context into the software that they use.

The new reality: If you have the appropriate schema to support customer workflows, the right administrative capabilities, and proper agentic interfaces, why the hell should a customer spend $100K on a systems integrator rolling kids off the bus to "learn your business" and deploy software that should already understand their business?

If customers are buying and deploying your software to solve core business processes, have your own deployment agent ask them questions (or better yet, summarize what you learned from the entire sales discovery process), recommend a setup and configuration, and just deploy it. When you release new capabilities, let them know what you can do for them now, whether they'd like to use it, and if so, simply deploy it and guide them on how you can be even more helpful.

Moving beyond setup and configuration, extra credit goes to companies that do what will become table stakes very soon: help customers set up workflows, business processes, and the associated “analytics” by recommending standard ones for companies like them, review existing processes and identify what's suboptimal, or connect them to process experts with portable technical context.

The Contextualized Software Revolution

After meeting with numerous existing and prospective founders over the past few months who are working at the cutting edge of what’s available with AI, I have come to the conclusion that we are witnessing the demise of "one-size-fits-all" software and the emergence of infinitely personalized applications. Instead of customers molding their businesses to fit rigid software schemas, AI-powered platforms will mold themselves to fit each customer's unique context. This isn't just customization—it's software that understands your industry, analyzes your existing processes, learns from your data, and dynamically generates the exact schema, workflows, and interfaces you need. A construction company and a law firm buying the same "project management" platform will receive completely different applications—one optimized for job sites, permits, and subcontractor coordination, the other for case timelines, document management, and billing cycles.

Imagine a world where:

Dynamic Schema Generation
Software companies don't offer customers a fixed schema. Instead, they analyze the customer's website, SIC codes, documents, and existing systems—then recommend the right underlying schema for them on the fly.

AI-Driven Process Optimization
In addition to the custom schema, the software recommends business processes that reflect what the customer does and best practices from similar companies. Users can only implement processes based on their existing knowledge. But what if the software knows better processes and can recommend them?

Predictive Analytics and Auto-Optimization
Instead of users setting up dashboards for questions they think they should ask, the software recommends the top 5-10 questions they should be asking. It provides answers and asks for permission to fix underlying processes or send instructions to other systems: "Your ROAS on this campaign is underperforming because of X. We think you should change it with these parameters. With your approval, we'll make those changes in real-time."

On-Demand Feature Creation
If you want your software to do something new for you—such as configure and price quotes—instead of going to an app store and evaluating 10 different CPQ solutions through a multi-week evaluation process, you simply tell your software, "I want you to do quoting for me." It already has a history of whatever documents you've been using for quotes, the context around products, pricing, and terms, and can build for you a quoting app that works exactly the way you work. No external integrations. No new vendor relationships. No learning curve.

Purpose-Built Interfaces
Imagine user interfaces automatically generated that are most purpose-suited to do the job and understand their performance. This is the farthest away from what’s in the market today, but everyone agrees that chat interfaces intermixed with tabs, views, and tables won't be the long-term winner. Instead of navigating predetermined screens designed by product managers who've never done your job, AI will dynamically create the perfect interface for each task. Approving invoices achieves a streamlined process with only the relevant data points. Managing construction projects requires real-time interfaces combining site photos, weather, and schedules. The interface becomes as contextual as the data—fluid, adaptive, optimized for how you think.

This world doesn't just exist in theory—it's here today for "smaller" or "less commercial" apps. Look at what founders are building with Lovable, Airtable's AI capabilities, and the myriad of AI-powered app-building platforms. And it's coming for every category of application software. Soon.

When it does, the entire framework of horizontal vs. vertical software collapses in that these three choices below are not the only ones:

Horizontal Software → One size of capability for all different types of customers
Vertical Editions → One size of slightly tweaked capability for more specific customer types
Vertical Software → A set of stacked capabilities for vertical-specific customers

Instead, what emerges is:

Contextualized Software → A set of personalized capabilities tailored for exactly what you need and want to do

We're moving from "one solution for everyone" to "a very customized solution out of the box for every specific customer."

Here's the catch: to deliver truly contextualized software, companies and founders must start within a given domain to effectively translate potential customer needs into the appropriate schema and workflows, then expand from there.

You can't build contextualized software for "everyone." You need deep domain expertise to understand what schemas actually matter for specific business types, which processes drive real value (as opposed to what customers think they need), how to translate business context into technical implementation, and what questions customers should be asking but aren't. This is obviously much easier to do with a vertical industry focus, but what should be clear is that this contextual software revolution might actually level the playing field somewhat for horizontal companies that can execute like champions.

Implications for Founders and the Industry

1. Implementation and the SI Industry Extinction

Difficult onboarding and implementation are on the way out—sorry, it just is. Every SaaS application will have its own implementation and onboarding agent that quickly guides customers through deployment and recommends areas where the software is not being utilized effectively. The systems integrator industry is completely at risk unless it evolves to become experts in providing context around business processes that customers can easily license and deploy. The "we'll learn your business and configure software for you" model is dead when AI can do it in hours instead of months.

2. System of Records Without Actions = Irrelevance

Applications that don't offer a powerful system of action layer are going extinct. If you don't make it easier for customers to get valuable work done better and smarter than before, what value does your system of record metadata matter? Data without action is just expensive storage.

3. Systems of Actions Can Eat Systems of Records

Here's the existential threat that systems of records players don't see coming: systems of actions can actually become systems of records if need be. It's simply not that hard to create a schema, especially a fixed schema, and as such, if systems of record players are playing hardball, it's much easier to go and create that schema with minimal R&D work.

Most systems of actions are already integrating with systems of records and leveraging that data appropriately in all the ways users care about. Over time, when the system of action says, "I can replace that system of record," users will be like, "What system of record?" They won't even remember why they needed the original database when their action-oriented software handles everything seamlessly. This is the ultimate checkmate move—systems of actions can selectively absorb the valuable parts of systems of records while discarding the bloat.

4. Horizontal Software: Easier & Harder

It’s harder for horizontal software companies to achieve breakout success unless you start within one core domain and quickly offer broad schema extension/building on-the-fly around adjacent processes. But it’s now actually easier to move from a singular horizontal solution to an extended horizontal solution as you quickly personalize apps for customers and chart a faster path to a compound startup. If customers don't want new tabs and screens, then building a configuration to solve vertical needs doesn't require new vertical editions—just knowledge of industry-specific adjustments that can be ingested during initial schema building and workflow recommendations.

5. Vertical Software: Advantage Amplified or Lost

Vertical companies have unique GTM advantages and AI context advantages given their laser focus on specific jobs-to-be-done. However, they need to double down and lean heavily into frictionless setup with flexible schema building, intentionally recommended business processes, superior agent-based analytics that drive actions (not just diagnosis), and rapid expansion of new jobs-to-be-done for customers.

If they don't lean all-in on their contextual advantage, they'll be replaced by either new vertical companies or horizontal companies with vertical context derived from their customers.

6. When Reinforcement Learning Comes - Pounce

The ultimate implication is still coming, but it's the most powerful one: when reinforcement learning truly arrives, the winners will have something far more valuable than proprietary data—they'll have proprietary know-how. Over time, the more jobs you do for a given type of company and the more you recommend business processes that work versus those that don't, you'll build software that comes to each customer with accumulated intelligence. Not just knowing how to set up the right schema and workflows for them, but more importantly, knowing how to help them run their business better. Your software won't just understand their context—it will understand what actually drives success in their industry based on thousands of other implementations. This creates the ultimate moat: software that gets smarter about business strategy with every customer deployment. While competitors are still figuring out schemas, you're delivering battle-tested playbooks for operational excellence. And that is powerful.