Over the last few weeks, three articles have forced me to rethink everything I thought I knew about Enterprise IT:

Jaya Gupta (Foundation Capital) on Context Graphs - arguing that the next trillion-dollar platforms will capture not just what happened, but why it was allowed to happen, creating systems of record for decisions.

Gokul Rajaram’s contrarian challenge to vertical AI - arguing that vertical AI founders face an existential threat from long-horizon agents like Claude Code that can work for hours or days on end, self-correct, and actually do stuff, without being trained on specific domains.

Aaron Levie (Box CEO) on the future of Enterprise IT - his conviction that current enterprise tools aren’t designed for AI agents, that we’ll have 100-1000x more agents than people, and that this opens a rare 15-year window for startups to disrupt incumbents with agent-first architectures.

What connects these perspectives? They’re all describing the same phenomenon from different angles: the fundamental physics of Enterprise IT is changing.

Just as physics describes the laws governing how matter and energy interact in the universe, Enterprise IT operates under its own set of “physical laws” that determine how data flows, how systems integrate, how value is created, and how complexity scales.

What we’re witnessing isn’t a technology upgrade - it’s a fundamental rewrite of these laws.

Part I: The Old Physics Is Breaking

The Five Laws of Current Enterprise IT

Law 1: The Law of Linear Complexity - O(N²)

Current Physics: Complexity grows quadratically with scale.

The formula is brutal: Integration Points = N × (N-1) / 2

With 250 applications, enterprises face 31,125 potential integration points (I know that not everything needs to connect to everything but pleae bear with me). Each new app doesn’t add linearly to complexity - it multiplies it.

A Fortune 500 company I spoke with spends upwards of $10M annually just maintaining these connections. Not building anything new. Just keeping the existing spaghetti from collapsing.

Governing Force: Point-to-point connectivity

Why it’s breaking: Organizations have hit the physical limit. At 250+ apps, no amount of governance, integration platforms, or middleware can manage this complexity. The system has collapsed under its own weight.

Law 2: The Law of Data Gravity

Current Physics: Data at rest tends to stay at rest (in silos).

Once data lands in an application database, it becomes “heavy” - nearly impossible to move. A major retailer has 47 different definitions of “customer” across their systems. Each creates its own gravitational well, pulling processes and workflows into isolated orbits.

Moving data between these wells requires massive “energy”: ETL jobs that run overnight, custom APIs that break with every update, data warehouses that replicate everything, and armies of engineers to manage it all.

Governing Force: Application-database coupling

What Jaya Gupta identified: This is worse than we thought. It’s not just that data is siloed - the decision context is missing entirely.

Decision traces capture what happened in this specific case: we used X definition, under policy v3.2, with a VP exception, based on precedent Z, and here’s what we changed. Agents don’t just need rules, they need access to the decision traces that show how rules were applied in the past, where exceptions were granted, how conflicts were resolved.

The synthesis that happened in someone’s head during that Slack thread, that Zoom call, that hallway conversation? Lost forever.

Law 3: The Law of User Friction

Current Physics: Work = Force × Distance across applications

Users must apply “force” (cognitive effort) to traverse “distance” (app switching). The average knowledge worker switches apps 10+ times per hour, losing 15 minutes of productivity with each context switch.

A sales rep closing a deal navigates: Salesforce (opportunity), LinkedIn (research), Gmail (communication), DocuSign (contracts), Slack (internal coordination), Zendesk (support history), billing system (pricing), approval workflow (discounts).

Eight different systems. Eight different logins. Eight different UI paradigms. Eight opportunities to lose context.

Governing Force: UI-mediated interaction

Aaron Levie’s observation: Generally, once you have a business process, you want to be able to define that in business logic with deterministic systems - just because the risk of that changing any given day is very high.

Current software forces humans to be the “integration layer” between disconnected systems. We are the universal adapters, translating data formats and business logic in our heads as we jump between tools.

Law 4: The Conservation of Integration Budget

Current Physics: 30-40% of IT budget permanently locked into maintaining connections.

Custom integrations create “debt” that must be paid indefinitely. When Salesforce updates their API, 50 downstream integrations break. When SAP releases a patch, integration testing takes weeks.

Innovation starves while “keeping the lights on” consumes resources.

A $10M IT budget? Only $6M is available for strategic initiatives. The rest is locked in integration maintenance - forever.

Governing Force: Proprietary APIs and tight coupling

Why this is unsustainable: As Aaron Levie points out, we’ll have about 100 times more, maybe 1,000 times more, agents than we have people. So you’ll have way more users of that software system, or SaaS, as agents.

The old physics - where integration scales with the number of systems - cannot support a world where agents outnumber humans 100:1. The math simply doesn’t work.

Law 5: The Entropy of SaaS Sprawl

Current Physics: Systems tend toward maximum disorder.

Left unchecked, application count always increases. Shadow IT accelerates this. Departments discover they have 347 SaaS subscriptions when IT thought they had 180.

48% of enterprise applications are unmanaged - dark matter in the enterprise IT universe. Unknown, ungoverned, unpatched, unsecured.

Governing Force: Departmental autonomy and vendor proliferation

The breaking point: Organizations attempting to deploy AI agents across this chaos discover it’s impossible. As Jaya Gupta notes, teams are hitting a wall that governance alone can’t solve.

Part II: The New Physics Emerging

The Critical Force: AI as the Catalyst

In the old physics, adding AI to existing architecture was like trying to add electricity to a steam engine. You could attach motors, but the fundamental mechanics remained unchanged.

In the new physics, AI is the electromagnetic force that reorganizes the entire system.

Here’s where Gokul Rajaram’s perspective becomes crucial. While Jaya Gupta argues that vertical, context-rich agents will win by capturing decision traces in the execution path, Gokul sees an existential challenge: long-horizon agents.

Agents that are not trained on a specific domain, but can reliably work for hours or days on end in pursuit of a goal, self-correct, and actually do stuff.

This is the key tension in the new physics:

• Jaya’s thesis: Vertical AI agents win because they sit in the execution path, capture decision traces, and train on enterprise-specific context

• Gokul’s counter: Long-horizon agents like Claude Code pose an existential challenge to vertical AI founders who’ve spent 2+ years building agents, training models on customer data, and embedding into workflows

The insight: While vertical AI companies build deep, trained context for specific domains, horizontal platforms are developing agents that can persist for hours or days, learning through execution rather than pre-training. Product engineering teams are being reduced by 50% while increasing product velocity, due to tools like Cursor and Claude Code, demonstrating the power of general-purpose agents that improve through use.

My synthesis: Both are right, but they’re describing different layers of the stack and different timescales for achieving dominance.

The Five New Laws

Law 1: The Law of Logarithmic Complexity - O(log N)

New Physics: Complexity grows logarithmically with orchestrated platforms.

Instead of every system connecting to every other system (N²), all systems connect to an orchestration layer. The orchestrator handles routing, translation, and coordination.

Adding the 251st capability doesn’t require 250 new integrations - just one connection to the orchestrator.

Formula: Integration Points = log(N) through orchestration layer

New Governing Force: Centralized AI orchestration

Aaron Levie’s architecture: The SaaS is used for the core business workflow, and then the agents ride on top of that. These agents would be helpful with making decisions, automating workflows, or essentially just accelerating whatever process the person was trying to do in the system.

This separation of concerns - deterministic core + non-deterministic agent layer - enables the logarithmic scaling. The orchestration layer manages the N² problem so individual systems don’t have to.

Law 2: The Law of Data Fluidity

New Physics: Data in motion stays in motion (frictionless flow).

Data exists in a unified fabric, accessible from anywhere. The semantic layer provides instant “translation” across contexts. Single source of truth eliminates gravitational wells.

New Governing Force: Data fabric and knowledge graphs

Jaya Gupta’s Context Graph: We call the accumulated structure formed by those traces a context graph: not the model’s chain-of-thought, but a living record of decision traces stitched across entities and time so precedent becomes searchable.

This is the missing piece. Current data architectures capture what happened (objects, events, transactions). Context graphs capture why it was allowed to happen (exceptions, overrides, precedents, cross-system context).

The flywheel: More workflows mediated → more traces captured → better at automating edge cases → more workflows automated. Data fluidity enables this compounding effect.

Law 3: The Law of Zero UI Friction

New Physics: Work = Intent (distance approaches zero)

Users express intent in natural language. AI agents traverse all system “distance” automatically. Context is maintained by agent memory.

Example transformation:

Old Physics:

• Open Jira → Create ticket

• Open Slack → Notify team

• Open GitHub → Create branch

• Open Confluence → Update docs

• (6 apps, 15 minutes)

New Physics:

• “Create a bug ticket for the login issue and notify the team”

• Agent handles everything

• (1 command, 30 seconds)

New Governing Force: Natural language interfaces and agent delegation

Aaron Levie on the work transformation: There’s a type of work that we never get around to in our company. I want AI to go and do that, because finally, it’s affordable for me to deploy AI agents at the kind of work that we could not fund before.

Example: “I’m sitting on 50,000 customer contracts. What if I could have an AI agent go through all those contracts and figure out which customers have the highest propensity to buy this next product?” This wasn’t feasible with human labor. It becomes trivial with agents.

Law 4: The Abundance of Innovation Budget

New Physics: Integration investment decreases over time.

Standard APIs and event-driven architecture eliminate custom work. Integration becomes configuration, not coding. Platform effects compound - each new capability makes others more valuable.

Result: 70%+ budget freed for innovation

New Governing Force: Event-driven architecture and standard protocols

The long-horizon agent advantage: Rather than spending months building perfect integrations upfront, agents work persistently toward goals. Like Claude Code - agents that can reliably work for hours or days on end in pursuit of a goal, self-correct, and actually do stuff.

This is radically more efficient than the old physics, where every integration required:

• Requirements gathering (weeks)

• Custom development (months)

• Testing (weeks)

• Ongoing maintenance (forever)

In the new physics: Deploy agent with a goal, let it work for hours or days to achieve it, move on.

Law 5: The Anti-Entropy of Platform Gravity

New Physics: Systems tend toward consolidation and order.

Unlike the old physics where inertia favored proliferation, the new physics creates strong forces toward consolidation.

Why? AI agents require unified data architecture. This creates evolutionary pressure.

New Governing Force: AI orchestration requirements and platform economics

Aaron Levie’s prediction: We are in this window right now that we have not been in for about 15 years, which is - there’s a complete platform shift happening in tech that’s opening up a spot for a new set of companies to emerge.

The opportunity: Startups can build “agent-first” from day one. Smaller startups have no business processes to change, so they can design a new process in an agent-first way.

Incumbents face the innovator’s dilemma: Cannibalize per-seat revenue or get disrupted. As Levie notes, the typical per-seat business model would no longer work, and instead companies would have to sell some form of consumption and volume-oriented use cases.

The changing physics of Enterprise IT isn’t coming. It’s here.

The question isn’t whether to transform.

It’s whether you’ll lead the transformation - or be transformed by it.

References

https://foundationcapital.com/context-graphs-ais-trillion-dollar-opportunity/