All-In's Sacks on the SaaS-to-Agents Transition

The Shift Nobody's Talking About Yet

David Sacks-co-founder of Slack, former Yammer CEO, and one of the All-In Podcast's sharpest voices on founder dynamics-recently laid out a thesis that should matter to every SaaS founder, operator, and investor watching the AI wave: not all software is equal when it comes to agent disruption.

The conventional take is that AI agents will "replace" SaaS. That's lazy. The real story is messier, more specific, and frankly more interesting. Some SaaS categories face existential pressure from autonomous agents. Others will become the infrastructure that agents run on. And some will barely notice the transition at all.

Sacks's argument cuts through the noise because he's not a pure AI evangelist or a SaaS purist-he's an operator who's built, scaled, and invested in both. His All-In segment on SaaS versus agents isn't a manifesto; it's a taxonomy. And for founders raising capital in 2025, understanding that taxonomy is the difference between a defensible business and a stranded asset.

Let's unpack what he actually said, why it matters, and which categories are most exposed.

Understanding the Agent-First Architecture Shift

Before diving into which SaaS plays are vulnerable, we need to establish what we mean by "agents" in this context. An AI agent isn't just a chatbot or a copilot bolted onto existing software. According to Deloitte's 2026 predictions on SaaS and AI agents, true agents represent a fundamental shift in how enterprise software operates-from tools that require human direction to systems that autonomously execute work within defined guardrails.

The architecture difference is critical. Traditional SaaS follows a user-centric model: a person logs in, navigates a UI, makes decisions, and triggers actions. The software amplifies human judgment but remains subordinate to it. Agents flip this. They observe context, reason about options, and execute decisions-with humans in an oversight role rather than a control role.

Sacks's point is that this architectural shift doesn't affect all SaaS equally. The transition from user-centric to agentic-centric systems creates winners and losers based on the nature of the work being automated. If your SaaS product is essentially a workflow coordinator-a system that collects inputs, applies rules, and produces outputs-then agents threaten your core value proposition. If your product is a data repository, a collaboration platform, or a specialized intelligence layer, agents might actually increase your defensibility.

This is where the real insight lives. And it's why we should care about what David Sacks and his All-In co-hosts are actually advising founders about in 2025-because their investment theses are shifting in real time.

The High-Exposure Categories: Workflow Automation and Coordination

Let's start with the most vulnerable segment: SaaS products built primarily to coordinate workflows and enforce business processes.

Think expense management tools. An employee submits an expense report. The system routes it to a manager. The manager reviews and approves or rejects. The system logs the decision and updates accounting systems. This entire flow can be executed by an agent. The agent reads the expense, pulls policy rules, checks budget constraints, contacts the manager if needed, and processes the approval. The human becomes an exception handler, not the primary operator.

Deloitte's analysis of agentic AI transformation identifies exactly this type of workflow as a primary target for agent disruption. Companies like Expensify, Concur, and Coupa face pressure not because agents are better at processing expenses (though they might be), but because the workflow no longer requires a dedicated SaaS interface. An agent can live inside Slack, email, or any communication layer and handle the entire process.

The same logic applies to:

• Approval routing and document workflows: Docusign, Adobe Sign, and similar e-signature platforms coordinate signature collection and approval chains. Agents can do this asynchronously, pulling documents, tracking status, and routing without a dedicated UI.
• Scheduling and calendar coordination: Calendly and similar tools solve for the problem of finding meeting times. An agent with access to calendars can do this natively, without a separate interface.
• Invoice and billing coordination: Bill.com, Stripe Billing, and similar platforms manage the back-and-forth of invoicing, payment, and reconciliation. Agents can orchestrate this workflow across multiple systems.
• Basic customer support triage: Zendesk and Intercom are increasingly under pressure from agents that can read support tickets, classify them, draft responses, and escalate to humans only when needed.

The pattern is clear: if your SaaS product's primary value is moving information and decisions through a defined process, agents are a direct threat.

But here's the nuance Sacks emphasizes: the timeline matters enormously. Agents capable of handling truly complex, multi-step workflows with high stakes are still maturing. An agent that can process a simple expense report? That's here now. An agent that can negotiate a complex contract or manage a multi-month procurement process with dozens of stakeholders? That's still a few years out. This means high-exposure SaaS companies have a window to either (1) become infrastructure that agents run on, or (2) pivot their value prop away from workflow coordination.

The Medium-Exposure Categories: Data Aggregation and Reporting

Next tier down: SaaS products that primarily aggregate data, generate reports, and provide visibility into business processes.

This includes business intelligence tools, analytics platforms, and operational dashboards. Tableau, Looker, Sisense, and similar BI tools face meaningful pressure from agents that can query data, generate insights, and surface anomalies without requiring a human to log in and build a dashboard.

However-and this is important-the exposure here is medium, not high. Why? Because the value isn't just the data; it's the structure and interpretation of the data. A BI tool provides a canonical source of truth. It enforces data governance, quality, and consistency. An agent can query that same data, but it doesn't replace the BI tool; it becomes a consumer of it.

McKinsey's analysis of generative AI and the future of work notes that the highest-value use cases for AI agents involve autonomous execution of well-defined tasks, not interpretation of ambiguous data. Reporting and analytics often involve judgment calls, context, and nuance that agents struggle with. A human still needs to ask the right questions.

The real risk for this category is commoditization, not displacement. If agents can generate 80% of the insights a BI tool provides, then BI tool pricing pressure increases dramatically. But the tool itself remains valuable as infrastructure.

Similarly, CRM systems (Salesforce, HubSpot, Pipedrive) face pressure on the usage side-agents can log activities, update deal stages, and track pipeline without a human doing data entry. But the CRM remains the system of record. The threat is that CRM vendors lose direct user engagement and must shift to a more API-first, agent-native model.

The Low-Exposure Categories: Data Repositories and Specialized Intelligence

Now the categories that are actually strengthened by the agent transition.

Data repositories-databases, data warehouses, data lakes-become more valuable in an agentic world, not less. Agents need reliable, clean, well-structured data to operate effectively. If your SaaS is Snowflake, BigQuery, or a specialized data platform, agents increase your TAM (total addressable market) because every agent needs a data source.

Similarly, specialized intelligence layers-products that provide domain expertise, proprietary data, or unique analytical capability-become more defensible. Consider a SaaS product that aggregates real estate market data, provides pricing models, and surfaces investment opportunities. An agent can consume that data and execute transactions, but it can't create the intelligence layer. The product becomes more valuable, not less, because agents can now act on those insights at scale.

How SaaS applications are moving from user-centric to agentic-centric architectures highlights that specialized SaaS products that provide unique intelligence or data will see their value increase, while commoditized workflow tools will face compression.

This also applies to:

• Compliance and risk management tools: If your SaaS provides regulatory expertise, risk models, or compliance intelligence, agents need you more than ever. They can execute on your guidance at scale.
• Vertical-specific software: A SaaS tool built for, say, veterinary clinics, that understands veterinary workflows, regulatory requirements, and best practices becomes more defensible, not less. Agents can operate within your domain framework.
• Collaboration and knowledge management: Slack, Notion, and similar tools become more central in an agentic world because agents need a place to operate and collaborate with humans. The shift is from a tool you use to a platform agents run on.

The Pricing and Unit Economics Compression

Beyond displacement, Sacks's analysis touches on a subtler but equally important pressure: unit economics compression.

Even if agents don't replace your SaaS product, they reduce the value per user. If an expense management tool processes 10x more expenses per human user because agents handle the routine cases, then the tool's revenue per user drops dramatically. This forces a pricing model shift-from per-user to per-transaction, from seat-based to consumption-based.

Deloitte's 2026 predictions note that agentic AI will drive gradual transformation of SaaS markets and enterprise spending patterns, which includes a shift toward consumption-based pricing and away from traditional seat licenses.

For founders raising capital, this is critical. If you're building a traditional SaaS product in a high-exposure category (workflow automation, approval routing, basic customer support), your path to a $100M+ exit just got narrower. Your TAM might actually shrink as agents handle more work per user. Your pricing power decreases. Your gross margins come under pressure because you're now competing with a commodity intelligence layer (the agent) rather than a specialized tool.

This doesn't mean these businesses are doomed. It means they need to evolve faster. The companies that survive are those that:

1. Become agent infrastructure: Shift from a user-facing tool to an API-first platform that agents consume. This is what Zapier is doing-positioning as the orchestration layer for agents, not the UI for humans.
2. Add specialized intelligence: Layer domain expertise, proprietary data, or unique workflows on top of the commodity agent layer. This is harder to replicate.
3. Move upstream or downstream: If workflow automation is commoditizing, move to the problem before the workflow (data collection, decision framing) or after it (outcome measurement, compliance verification).

What This Means for Founders Raising Capital

If you're a SaaS founder in a high-exposure category, here's what investors are thinking (and what Sacks would likely tell you directly):

Your market timing is compressed. You have maybe 18-36 months before agents become sufficiently capable that your core workflow starts to look commoditized. This means your path to profitability or a strong exit needs to accelerate. You're not raising a leisurely Series B to build features; you're raising to either dominate your niche or pivot your value prop.

Your defensibility story needs to evolve. If you're pitching "we make expense management easier," investors hear "you're a feature, not a business." If you're pitching "we're the infrastructure layer that agents depend on for reliable expense data and policy enforcement," that's a different conversation. Check out how other founders are adapting their capital raising strategies to see which narrative patterns resonate with institutional VCs.

Your pricing model might need to shift before you want it to. If you're planning to grow revenue per user over time, reconsider. Agents will compress that. Moving to consumption-based pricing (transactions processed, data queries, API calls) isn't a nice-to-have; it's a survival mechanism.

Vertical specialization becomes your moat. If you're a horizontal workflow tool, you're in trouble. If you're a workflow tool built specifically for, say, financial services compliance or healthcare credentialing, you have a defensible niche because agents still need domain expertise to operate safely.

For investors, the implication is equally clear. All-In's hosts have been tracking the AI investment thesis carefully, and Sacks's SaaS-to-agents segment signals a shift in how they're evaluating SaaS deals. High-exposure categories are getting lower valuations and shorter runways. Low-exposure categories-especially those positioned as agent infrastructure-are getting premium multiples.

The Timing Question: When Does This Actually Happen?

One of Sacks's most pragmatic points is that the agent transition isn't happening all at once. It's not like the cloud transition, where there was a clear before and after. It's more like the mobile transition-gradual, category-specific, and full of false starts.

Some categories will see agent disruption in 2025. Others won't feel it until 2027 or 2028. And some might never fully transition because the regulatory, security, or domain-specific complexity is too high.

For example:

• Expense management: Agents can handle this now. Pressure is immediate.
• Complex contract negotiation: Agents will struggle with this for years. Pressure is delayed.
• Customer support triage: Agents are capable now, but human oversight remains critical. Pressure is moderate and ongoing.

Gartner's 2024 hype cycle assessment of agentic AI places many agent use cases in the "slope of enlightenment" phase, meaning they're moving from hype to practical deployment, but not yet mainstream. This gives founders a window, but the window is closing faster than most realize.

The Infrastructure Play: Where Sacks Sees Real Opportunity

If Sacks's analysis has a silver lining for SaaS founders, it's this: the shift from user-centric to agent-centric software creates enormous infrastructure opportunities.

Every agent needs:

• Data infrastructure: Where does it pull context from? This is where data warehouses, data lakes, and specialized data platforms win.
• Orchestration infrastructure: How does it coordinate across multiple systems? This is where platforms like Zapier, Make, and n8n become more valuable.
• Governance and safety infrastructure: How do you ensure agents don't make mistakes or violate policies? This is where compliance, audit, and risk management tools become critical.
• Observability and monitoring: How do you know what your agents are doing? This is where logging, monitoring, and analytics tools become essential.

How SaaS is evolving toward agentic-centric models emphasizes that the infrastructure layer-the plumbing that agents run on-is where the defensible, high-margin businesses are being built.

Sacks's implicit thesis is that the next generation of $10B+ SaaS companies won't be user-facing workflow tools. They'll be the infrastructure that every agent in the enterprise depends on. If you're building SaaS in 2025, ask yourself: "Is this something agents will need to depend on, or something agents will replace?"

Real Numbers and Category Breakdown

Let's get specific. Here's how Sacks would likely categorize the major SaaS categories by exposure:

High Exposure (Timeline: 12-24 months)

• Expense management (Expensify, Concur, Divvy)
• Approval routing and workflow (Nintex, Appian)
• Basic customer support triage (Zendesk, Intercom for tier-1 issues)
• Calendar and scheduling (Calendly)
• Invoice and payment coordination (Bill.com, Stripe Billing)

Medium Exposure (Timeline: 24-36 months)

• CRM (Salesforce, HubSpot, Pipedrive)
• Business intelligence (Tableau, Looker, Sisense)
• Project management (Asana, Monday, Jira for routine task assignment)
• Email and communication (Gmail, Outlook, Slack for routine message routing)

Low Exposure (Timeline: 36+ months or indefinite)

• Data warehouses (Snowflake, BigQuery)
• Vertical-specific software (medical practice management, legal case management)
• Specialized intelligence platforms (real estate data, financial research)
• Collaboration and knowledge management (Notion, Confluence)
• Observability and monitoring (Datadog, New Relic)

What Founders Should Do Right Now

If you're raising capital for a SaaS startup in 2025, here's the Sacks-aligned playbook:

1. Honestly assess your exposure. Don't rationalize your way out of this. If your product is primarily workflow automation, you're in the high-exposure bucket. Own it.

2. Articulate your defensibility against agents. Is it domain expertise? Proprietary data? Network effects? Regulatory moats? Pick one and double down. "We have a great UI" is not a defensibility story anymore.

3. Consider your pricing model now. If you're planning to grow revenue per user, that's a bet against agent adoption. Consider shifting to consumption-based pricing before you're forced to.

4. Position as infrastructure, not a tool. Instead of "we make X easier," try "we're the system of record and governance layer for X, and every agent in the enterprise depends on us."

5. Accelerate your go-to-market. Your window to build a defensible moat is shorter than you think. If you're planning a leisurely Series B, reconsider. You might need to compress your timeline to Series C or profitability.

For context on how other founders are navigating this shift, check out the capital raising playbooks that are resonating with institutional VCs in 2025.

The Broader Implication: SaaS Isn't Dead, It's Evolving

Here's where Sacks's thesis gets most interesting: he's not saying SaaS is dead. He's saying the form of SaaS is changing. The companies that will dominate the next decade won't be user-facing workflow tools. They'll be the infrastructure layers, the specialized intelligence platforms, and the domain-specific tools that agents depend on.

Traditional SaaS is transitioning to AI-native SaaS with fundamentally different execution models, which means the playbook for building a $100M+ SaaS business is different now. You can't just execute the same playbook that worked for Salesforce or Slack. You need to build for an agentic world.

This is why understanding how the All-In hosts are actually advising founders matters. They're not just talking about this in podcasts; they're backing it with capital. If you're raising, you need to understand their investment theses and position accordingly.

The Bottom Line

Sacks's analysis of SaaS versus agents isn't a prediction that SaaS is dying. It's a taxonomy of which SaaS categories face disruption, which face compression, and which actually become more valuable. For founders, the implication is clear: if you're building in a high-exposure category, you need to evolve your value prop, accelerate your timeline, and shift your defensibility story. If you're building infrastructure or specialized intelligence, you're in a stronger position than ever.

The founders who survive the next 36 months are those who see this transition not as a threat to SaaS, but as a fundamental reshaping of what SaaS looks like. The winners will be those who position themselves as the infrastructure that agents depend on, not the tools that agents replace.

For more on how top investors are actually thinking about AI and SaaS in 2025, explore how David Sacks and other All-In hosts are positioning their portfolios and what their actual investment theses look like.