Click any layer to explore
Each node reveals what's changing, why it matters, and the core debate between bull and bear perspectives.
How to read the diagram:
▬ Purple arrows = AI agent's new direct paths to data and workflows
▬ Red dashed = disrupted / bypassed by AI
▬ Gray = traditional flow (user → UI → app → data)
➤ Key insight: AI agents bypass the UI layer and access data/workflows directly
AI agents represent a fundamentally new layer — they reason, plan, and execute multi-step tasks autonomously, sitting on top of existing software and potentially capturing the value that used to sit below.
Software vendors sold workflows for humans to follow. AI agents execute those workflows directly — the interface shifts from the vendor's UI to the agent's conversational layer. The software beneath becomes infrastructure, not the product.
Dashboards, forms, report builders, portals — this is where disruption hits hardest and fastest. When an AI agent can query data and present it conversationally, pre-built display layers lose their value.
Every bank, retailer, and services firm has invested heavily in customer portals and dashboards. The question: who captures the value when interactions move from structured interfaces to conversational AI?
Most exposed: Tier-1 customer support, basic BI/reporting, search-and-display apps, templated content management.
Timing debate: CIO surveys suggest 5-10 years before meaningful displacement. But the market is pricing it now — stocks down 20-50% on forward-looking fear alone.
Application vendors (Salesforce, ServiceNow, SAP, Workday) provide structured business logic. Their fate depends on whether they sell workflow orchestration (replaceable) or embedded operational context (defensible). AI agents "orchestrate" this layer — the app becomes middleware that agents call, not software humans operate.
Near-consensus across all research: proprietary data is the most defensible asset in an AI world. Agents are only as good as the data they operate on. The purple "queries directly" arrow is key — agents bypass apps to access data, making this layer more valuable.
The shift from "system of record" to "system of context" — making data meaningful for AI agents. A bank's transaction history, a staffing firm's placement database — these become the foundation agents depend on. Who owns this context layer for your industry?
For non-tech companies: Your internal data — customer records, transaction patterns, operational history — is increasingly your most strategic asset. Companies that invested in data governance are 2-3 years ahead.
Cloud providers are the clearest beneficiaries. AI workloads are compute-intensive — GenAI drives a 5-10x multiplier in inference spend vs. traditional cloud. Azure growing 38-39% YoY, supply-constrained.
Key tension: Capex intensity is extreme (Microsoft spending $38B/quarter, 46% of revenue). And concentration risk is real — Microsoft has $281B in contracted revenue from OpenAI alone. Multi-vendor strategies emerging as enterprises hedge.
Click a shift to explore
Each represents a structural break from how previous technology waves worked.
Previous shifts (cloud, mobile) added deployment channels but preserved per-seat subscriptions. AI challenges the model itself — agents replace human users, collapsing seat economics.
The transition from seat-based to consumption/outcome-based pricing is one of the hardest organizational shifts in enterprise software history. Companies that delay risk being undercut by AI-native alternatives with consumption models from day one.
AI coding tools have dropped the cost of building custom software dramatically. "Vibe coding" — where non-engineers describe what they want and get working software — lowers the barrier to internal development.
AI model capabilities improve on a timeline of months. Each improvement makes new categories of automation possible — creating a rolling wave of disruption across software verticals.
The competitive landscape can shift materially between quarterly earnings calls. Strategic plans designed for 3-year horizons may be obsolete within 12 months. Speed of execution — not strategy quality — is now the primary differentiator.
This requires flattening decision hierarchies, empowering teams to experiment without committee approval, and accepting higher failure rates. The cost of a failed AI experiment is small. The cost of being 12 months late to a structural shift can be existential.
Most incumbent software companies are bundles — proprietary data, workflow logic, UI, and network effects packaged at a single price. AI doesn't need to replace the whole bundle to destroy value — it just needs to make parts of it contestable, forcing the company to unbundle. The layers that were never truly defensible can no longer command premium pricing.
Bloomberg charges ~$25K/year per terminal, bundling: proprietary data, network effects (Bloomberg chat = Wall Street's messaging layer), workflow tools, and the terminal UI. The data and chat network are hard to displace — but the terminal interface and analytics workflows are not. AI agents that query Bloomberg data directly put pressure on the bundled price even if the core data moat holds. The company survives, but at what margin?
The key question for every incumbent: if customers could pay separately for each component, which pieces would they still buy at current prices? The gap between the bundled price and the sum of defensible parts is the value at risk.
Applies broadly: Salesforce bundles CRM data + pipeline UI + reporting. SAP bundles ERP logic + dashboards + compliance data. In each case, AI is making it easier for customers to see which pieces are defensible and which are not.
Growth for all — TAM expands
Existential threat
Incumbents win — AI is a feature
Self-cannibalize or be disrupted
Click a quadrant to explore
Each scenario has distinct implications for which companies survive, adapt, or face existential risk.
AI doesn't replace the software — it creates dramatically more demand. Every agent needs data infrastructure, security governance, and observability. TAM expands from serving humans to serving humans + agents.
Examples: Snowflake/Databricks (data foundation for agents), Datadog (observability of agent behavior), Okta/CrowdStrike (agent auth is unsolved), MongoDB (agentic memory + state).
The value proposition — "we help humans do X" — collapses when AI does X better, faster, and cheaper without a human in the loop.
Most exposed: Tier-1 support (AI handles 60-80% of inquiries), content generation (commoditized by LLMs), simple workflow automation (anyone builds with Claude Code), translation (real-time translation threatens language learning).
Timing caveat: CIO surveys say 5-10 years to full displacement. But stock markets price forward — hence the immediate valuation collapse.
Where workflows are complex, regulated, or deeply embedded, AI becomes a feature within existing software rather than a replacement.
Examples: Intuit (20 LLMs, tax code complexity), Autodesk (BIM data standard), SAP/Oracle ERP (decades of customized logic), vertical AI with FDA/legal clearance.
Workflows can be automated, but AI hasn't fully penetrated yet — a window of opportunity and danger. Companies must choose between protecting existing revenue and cannibalizing themselves.
Newspaper earnings grew 2002-2007, but stock prices fell in a straight line as the market priced Craigslist's structural threat. The same dynamic is playing out in SaaS today.
Examples: Salesforce (Agentforce vs. CRM automation), Workday (HR automation vs. system-of-record stickiness), expense/travel tools (agents can do it end-to-end).
Click a factor to explore
Ranked from strongest to weakest defensibility in an AI-disrupted world.
Hamilton Helmer's "7 Powers" framework remains the best lens for durable competitive advantage — even in an AI world. The key question: which of these moats survive when AI commoditizes the software layer?
Key insight: The top 3 powers (network effects, switching costs, scale) remain strong in AI. The bottom 2 (counter-positioning, process power) may actually favor AI-native challengers.
Companies whose core value is proprietary data that improves with usage have the strongest moat. AI makes this data more valuable — agents are only as good as the data they operate on. The flywheel: more users/agents → better data → better AI → more users.
Think: Palantir (ontology mapping), Intuit (tax data), Bloomberg (the proprietary datasets clients can't get elsewhere). For non-tech companies: customer transaction data, operational history, and domain knowledge fit here.
Software that actually executes — processing payments, filing regulatory documents, managing supply chains — is harder to displace. Agents need reliable execution paths.
Key distinction: "Read path" software (dashboards, reports, search) = highly vulnerable. "Write path" software (transactions, compliance enforcement, workflow execution) = defensible. Ask: does this software do things, or show things?
Being the authoritative source of truth provides switching cost protection. But there's a real debate about whether this alone is enough.
Software whose primary value is organizing and displaying information faces the most direct disruption. AI agents query underlying data directly and present it however the user needs.
The uncomfortable question: If I could just ask an AI agent to get me this information, would I still open this application? If "probably not," the product is in the danger zone.
Most software companies are beating earnings estimates and generating strong free cash flow. Yet stocks are falling 20-50%. The market is not reacting to current results — it is repricing the terminal value of these businesses based on AI disruption risk 3-5+ years out. This mirrors how newspapers performed in the 2000s: earnings grew 2002-2007, but stocks fell in a straight line as the market priced the internet's structural threat.
Click a force to explore
Understanding the mechanics behind the valuation reset.
Software stocks are priced on DCFs where 60-80% of valuation comes from "terminal value" — earnings expected beyond 5-10 years. When AI creates uncertainty about whether a business will exist in its current form, the terminal value collapses even if near-term earnings are strong.
Uncertainty alone kills valuations. The market doesn't need to believe SaaS is dead — it just needs to believe the range of outcomes has widened. When the downside includes "this category may not exist in 10 years," discount rates rise and multiples compress.
The median forward revenue multiple has dropped from 11x (2021) to 3.4x. This isn't about current performance — it's a permanent re-rating of the durability premium these businesses once commanded.
While most SaaS companies face future disruption risk, several categories are seeing AI impact on earnings today:
Every SaaS company faces a fundamental strategic choice:
The market's verdict: Most companies are stuck between paths. Growth decelerated from 44% (2021) to 18% (2025), net retention declining from 124% to 109%. The market is penalizing indecision.
Click a question to explore
Use these questions to guide a conversation with your leadership team about AI readiness.
The market calibrates on leadership conviction. Companies whose CEOs signal AI as a bolt-on feature get penalized; those who signal fundamental reinvention get rewarded.
AI model capabilities improve on a timeline of months. Speed of execution — not strategy quality — is now the primary differentiator.
In an AI world, software is commodity; data is defensibility. Do you have data that no one else has, and does it improve as more people/agents use your system?
CIOs flag data quality as the #1 blocker to AI adoption. Companies with clean, governed data are 2-3 years ahead.
Can your organization cannibalize its own revenue streams before competitors do? This requires leadership that protects innovators from the antibodies of the existing business.
Senior AI engineers command $500K+. If you can't compete for this talent, your AI strategy is aspirational, not executable.
The biggest trap is "AI add-on." Don't add AI to your expense tool — ask whether expense reporting should exist as a human activity at all.
Click an action to explore
The five things every leadership team must get right.
The single most common mistake: treating AI as a feature to bolt onto existing products. The companies that win ask "should this entire workflow exist?" — not "how do we make it 10% faster?"
The test: For each major process, ask: "If we were building this company from scratch today, would this process exist in its current form?" If no, you need a plan to change it — before someone else builds the alternative.
Across all sources — VC, sell-side, and operators — the consensus is clear: proprietary data is the only durable moat. Software can be replicated. A bank's 20 years of transaction patterns cannot.
Priority actions: Consolidate data infrastructure. Invest in data quality and governance. Break down silos. Build APIs that make data accessible to AI agents. Without this foundation, every other AI initiative underperforms.
The competitive landscape can shift materially between quarterly earnings calls. Companies that spend 6 months perfecting an AI strategy will find the landscape has changed by the time they execute.
Implication: Flatten decision hierarchies, empower teams to experiment without committee approval, accept higher failure rates. The cost of a failed AI experiment is small. The cost of being 12 months late can be existential.
Disruption is not a one-time event — it's a continuous process that will accelerate. The organizations that survive institutionalize change, not execute one successful transformation.
The test: Does your most innovative team have air cover from the CEO? Can they pursue projects that threaten existing revenue without being shut down by the business unit that owns it? If not, your innovation is performative.
AI is restructuring the boundary between human and machine work. Some roles face 30-50% reduction (legal paralegals, tier-1 support). Others become 10x more productive (engineers, analysts). The net effect is reallocation, not simple headcount reduction.
The opportunity: Frame AI as "freeing our best people to do higher-value work" and you retain talent. Frame it as "cutting costs" and you lose the people you most need. The CEO's framing matters for morale and market perception alike.