Peter Fenton's Enterprise Bets in the Agent Wave

The Pattern Nobody Talks About

Peter Fenton has backed some of the most consequential enterprise software companies of the last two decades-Yelp, Twitter, Zendesk, New Relic, Dropbox. The list reads like a highlight reel of exits and IPOs. But there's a deeper pattern running through his portfolio that matters more than any single success: a consistent thesis about how enterprises fundamentally change when new tools arrive.

That thesis is now playing out in real time with AI agents.

Fenton isn't a flashy venture investor. He doesn't chase headlines or declare that everything will be "disrupted." Instead, he sits at Benchmark Capital and makes deliberate bets on companies that solve a specific kind of enterprise problem: the friction points that emerge when incumbents can't adapt fast enough to new technology. His track record suggests he understands something that most investors miss-that the biggest wins come not from building entirely new categories, but from identifying where existing workflows are about to break and positioning capital behind the teams that can rebuild them.

The agent wave is different, but the underlying logic is the same.

Understanding Enterprise AI Agents

Before dissecting Fenton's thesis, we need to be precise about what enterprise AI agents actually are. An agent, in the technical sense, is an AI system that can perceive its environment, make decisions, take actions, and iterate based on feedback-all with minimal human intervention. Unlike a chatbot that responds to user queries, an agent operates autonomously within defined parameters.

In enterprise contexts, this means agents can:

• Execute multi-step workflows across systems (CRM, ERP, databases, APIs) without human handoff
• Make judgment calls based on business rules and context
• Learn from outcomes and improve performance over time
• Operate 24/7 without fatigue or context-switching costs

The distinction matters because it changes the economic calculus of software. Traditional enterprise SaaS optimizes for user productivity-making humans faster. Agents optimize for task completion-replacing human decision-making altogether in specific domains.

This is what a16z identified as "The Agent Wave is Here"-a fundamental shift in how enterprise software creates value. And it's precisely the kind of shift that Fenton has built his career around spotting.

Fenton's Enterprise Thesis: The Disruption Pattern

When you examine Fenton's best bets across two decades, a thesis emerges. He's not investing in companies that "disrupt" in the abstract sense. He's investing in companies that solve a specific problem: how enterprises operate when the old way stops working.

Take Dropbox. When Dropbox launched, enterprise file sharing existed. Companies used email, FTP, and on-premises servers. Dropbox didn't invent file sharing-it made it frictionless. The enterprise was still managing files the same way; Dropbox just removed the friction points. Fenton's early backing of Dropbox reflected his understanding that enterprises would adopt tools that made existing workflows dramatically easier, even if those workflows were conceptually unchanged.

Zendesk followed the same pattern. Customer support existed before Zendesk. But the old way-scattered tickets, fragmented systems, manual routing-was breaking under scale. Zendesk didn't invent customer support; it rebuilt the workflow for a distributed, cloud-native world.

Yelp and Twitter were different categories entirely, but the thesis held. Both exploited a moment when existing institutions (restaurant guides, media distribution) couldn't adapt fast enough to new consumer behaviors. Fenton recognized that when institutions freeze, new platforms emerge to capture the value they're leaving on the table.

The pattern: Identify where incumbents are structurally unable to adapt. Back the team that rebuilds the workflow for the new era.

The Enterprise AI Agent Opportunity

Now apply that thesis to enterprise AI agents.

Every major enterprise today is running on legacy workflows that were designed for human decision-making. A sales rep qualifies leads manually. A support agent routes tickets by hand. A finance team manually reconciles invoices. A procurement officer approves purchases through email chains. These workflows are embedded in systems that are 10, 20, sometimes 30 years old.

AI agents don't optimize these workflows-they eliminate them.

This is where Fenton's thesis gets interesting. The incumbents (Salesforce, ServiceNow, SAP, Oracle) are trying to bolt AI onto existing products. They're adding agents as a feature. But they can't fundamentally rebuild their workflows because they're locked into legacy architecture, legacy pricing models, legacy sales motions, and legacy customer expectations.

Meanwhile, new companies can be born with agent-first architecture. No technical debt. No installed base of users expecting the old workflow. No sales team trained to sell humans productivity tools.

This is the moment Fenton has spent two decades recognizing. The moment when the old guard can't move fast enough, and new entrants can move at the speed of the new technology.

The McKinsey analysis on "The Enterprise AI Agent Revolution" frames this as a structural reshuffling of enterprise operations-exactly the kind of moment when Fenton's thesis applies.

Fenton's Actual Agent Bets

So what is Fenton actually backing in the agent wave?

Benchmark's portfolio page lists some obvious candidates-companies in infrastructure, data, and automation. But the more interesting signal comes from Fenton's public commentary and the pattern of Benchmark's recent investments.

Fenton has been explicit that the agent wave creates opportunities in three layers:

1. Agent Infrastructure & Orchestration

Companies that help enterprises build, deploy, and manage agents across their systems. This is the plumbing layer-the tools that let companies connect agents to their existing databases, APIs, and workflows. The play here is similar to Fenton's earlier infrastructure bets: identify the foundational layer that every agent deployment will need, and own that layer.

The economics are compelling. If every enterprise eventually runs hundreds of agents across their organization, the infrastructure that orchestrates those agents becomes as critical as databases or monitoring tools. Fenton backed New Relic in monitoring; he'd recognize the same pattern in agent orchestration.

2. Vertical AI Agents

Companies that build industry-specific agents for sales, support, finance, or operations. These aren't horizontal platforms-they're focused on solving one job in one industry with extreme depth.

This is where Fenton's thesis becomes most visible. Vertical agents require deep domain expertise, customer relationships, and workflow knowledge. A horizontal AI company can't build a world-class sales agent without understanding enterprise sales psychology. A startup can.

The Zendesk playbook applies here: own the workflow, own the customer relationship, own the outcome. A vertical sales agent company that can prove it increases qualified pipeline by 40% and reduces sales cycle time by 30% will command enterprise valuations.

3. Agent-Native Enterprise Platforms

Companies building the next generation of enterprise software from scratch, with agents as the core abstraction rather than users. These are the highest-risk, highest-reward bets-companies that could become the Salesforce or ServiceNow of the agent era.

Fenton would recognize this as the Dropbox moment-the chance to rebuild an entire category for a new paradigm.

The Valuation Thesis

Here's where understanding Fenton's approach matters for founders raising capital.

When Fenton backed Zendesk, he wasn't valuing it as a "customer support tool." He was valuing it as the infrastructure that would underpin how enterprises manage customer interactions in a distributed, cloud-native world. The valuation reflected not just the TAM of support software, but the structural inevitability that enterprises would need to rebuild their support operations.

The same logic applies to enterprise AI agents. A founder pitching an agent company shouldn't be pitching "we're 10% more efficient at lead qualification." They should be pitching "we're rebuilding the entire sales workflow for an era where agents handle 80% of qualification and humans handle 20% of judgment calls."

That's a different TAM. That's a different valuation. That's the difference between a $50M exit and a $5B exit.

For founders building in this space, understanding AI Startup Valuations: The Reality Check You Need for Fundraising Success is essential-but so is understanding the narrative framework Fenton uses to justify those valuations.

What Fenton Looks For in Agent Companies

Based on his track record, Fenton evaluates agent companies through a specific lens:

Founder-Investor Fit on Vision

Fenton doesn't back founders who are optimizing incrementally. He backs founders who have a vision for how an entire workflow will be rebuilt. When he backed Jack Dorsey at Twitter, he wasn't backing "a better way to post messages." He was backing a vision for how information distribution would fundamentally change.

For agent companies, this means: Do the founders understand not just the AI, but the enterprise workflow deeply enough to know what gets rebuilt vs. what stays human? Can they articulate why their agent architecture is fundamentally different from a chatbot bolted onto legacy software?

Defensibility Through Workflow Lock-in

Fenton's best bets had network effects or workflow lock-in. Dropbox became harder to leave the more files you stored. Yelp became more valuable the more reviews accumulated. Zendesk became more valuable the more tickets and customer context accumulated.

For agent companies, the equivalent is: What data does the agent accumulate that makes it harder to replace? A sales agent that learns your company's qualification criteria, your customer base, your deal patterns-that's locked in. A support agent that understands your product, your customers, and your escalation patterns-that's locked in.

Fenton would look for agent companies where the core value comes not just from the AI model, but from the proprietary data and workflow context the agent accumulates.

Go-to-Market Clarity

Fenton doesn't back companies with vague GTM. He backs companies with a clear wedge into the enterprise. Zendesk started with small teams and worked up. Dropbox started with engineers and worked across the org.

For agent companies, this means: What's the initial use case where the agent delivers undeniable value? Can you get to 10 customers, prove ROI, and expand from there? Or are you trying to sell "AI transformation" to the entire enterprise at once?

The best agent companies will likely follow a similar pattern-start with one workflow, prove the economics, then expand to adjacent workflows.

The Competitive Landscape

Fenton's approach to competition is instructive. He doesn't avoid crowded markets. He backs companies that will win in crowded markets because they understand the workflow better than anyone else.

In agent infrastructure, there's competition from Anthropic, OpenAI, and existing enterprise software vendors. But Fenton would back the company that understands how agents integrate with existing enterprise systems better than anyone else-the company that makes it trivially easy to deploy agents across a messy, heterogeneous tech stack.

In vertical agents, there's competition from both startups and incumbent vendors adding AI features. But Fenton would back the company that rebuilds the entire workflow, not just adds an AI feature.

This is where Andreessen Horowitz's $20B AI Fund: The 2025 Game Changer for U.S. Tech Startups matters-a16z is explicitly positioning for the agent wave, which means competition for the best deals is intensifying. Fenton's advantage is that he's been making these bets for 20 years. He knows how to spot the winner in a crowded field.

How Founders Should Pitch Agent Companies to Fenton

If you're a founder building an enterprise AI agent company and you want to get in front of Fenton, here's what matters:

1. Lead with the Workflow Insight

Don't start with "We built an AI agent that's 15% more efficient." Start with "We're rebuilding [sales qualification / customer support / invoice processing] for an era where humans handle judgment and agents handle execution. Here's why the old way breaks at scale."

Fenton wants to see that you understand the workflow deeply-not just the AI.

2. Show Founder-Investor Fit

Have you spent years in this workflow? Do you have deep relationships with customers? Can you articulate why you're the team to rebuild this, not a feature team at Salesforce or ServiceNow?

Fenton doesn't back AI experts building enterprise software. He backs enterprise software experts using AI as a tool.

3. Prove Early Unit Economics

Get to 3-5 customers and show that your agent delivers measurable ROI. Fenton doesn't need a massive TAM-he needs evidence that the workflow you're rebuilding is actually valuable enough that enterprises will pay for it.

For A Step-by-Step Guide for Entrepreneurs on How to Pitch Their AI Projects and Raise Private Money, the key is translating technical capability into business outcome. Fenton cares about outcomes.

4. Articulate the Expansion Path

Start with one workflow, but show how the agent can expand to adjacent workflows in the same customer. A sales agent that starts with qualification can expand to pipeline management, forecasting, and deal analysis. That's how you get from $1M ARR to $100M ARR.

Fenton wants to see that you're not building a point solution-you're building a platform that will eventually touch the entire enterprise.

The Macro Context: Why Now?

Fenton's timing on enterprise bets has always been tied to macro shifts. Dropbox came when cloud infrastructure became reliable enough to trust with files. Zendesk came when enterprises were moving to distributed teams and cloud-based systems. Twitter came when mobile devices made information distribution mobile-first.

The agent wave is happening now because three things converged:

1. Large Language Models Became Good Enough

GPT-4 and its competitors can handle nuanced enterprise workflows. They can reason about context, handle edge cases, and explain their decisions. This is new. Two years ago, AI agents were a research project. Now they're practical.

2. Enterprise APIs Became Standardized

Every major enterprise system (Salesforce, SAP, Oracle, Workday) has well-documented APIs. Agents can integrate with these systems reliably. This wasn't true five years ago.

3. Enterprises Are Under Pressure

Labor costs are rising. Competition is intensifying. Enterprises are desperate for ways to do more with less. An agent that can handle 80% of a workflow autonomously is suddenly worth a lot of money.

This is exactly the moment Fenton has spent his career waiting for. And it's exactly the moment when founders should be raising capital to build agent companies.

Understanding AI Gets 31% of Venture Funds in Q2, Q3 2024: A Deep Dive into the VC Landscape shows that the capital is flowing toward AI, but Fenton's thesis suggests that the capital will concentrate in companies that understand enterprise workflows deeply enough to rebuild them.

Learning from Fenton's Past Bets

Let's examine what made Fenton's previous bets successful and how those lessons apply to agents.

Yelp (2005): Fenton backed Yelp when online reviews were a new concept. The old way-asking friends, reading newspaper reviews-was slow and limited. Yelp didn't invent reviews; it made them crowdsourced and real-time. The defensibility came from the review data and the community that contributed it.

For agents: The equivalent is backing an agent company where the defensibility comes from the workflows the agent learns and the data it accumulates, not just the underlying AI model.

Dropbox (2006): Fenton backed Dropbox when file sharing was fragmented and painful. The old way-email attachments, FTP, USB drives-was clunky. Dropbox didn't invent file sharing; it made it seamless and device-agnostic. The defensibility came from the installed base and the switching costs.

For agents: The equivalent is backing an agent company where switching costs are high because the agent has learned your specific workflows, your customer base, and your business rules.

Zendesk (2009): Fenton backed Zendesk when customer support was stuck in an on-premises world. The old way-software you had to host yourself, limited integrations, siloed data-was breaking as companies went distributed. Zendesk didn't invent support software; it rebuilt it for the cloud era. The defensibility came from the customer data and the integrations that accumulated over time.

For agents: The equivalent is backing an agent company that rebuilds an entire workflow for the agent era, not just adds an agent feature to existing software.

New Relic (2008): Fenton backed New Relic when application monitoring was manual and reactive. The old way-log files, manual debugging, reactive incident response-was breaking as applications became more complex and distributed. New Relic didn't invent monitoring; it rebuilt it for the cloud era. The defensibility came from the instrumentation data and the customer lock-in.

For agents: The equivalent is backing an agent infrastructure company that becomes the standard layer that every enterprise depends on for orchestrating agents.

The pattern is consistent: Fenton backs companies that understand a workflow deeply enough to rebuild it for a new era, not companies that bolt new technology onto old workflows.

Practical Framework for Evaluating Agent Companies

If you're a founder, an operator, or an investor trying to understand which agent companies will win, here's the framework Fenton would likely use:

Workflow Understanding (40%)

• Do the founders have deep domain expertise in this workflow?
• Can they articulate why the old way breaks at scale?
• Do they have early customer relationships that validate the problem?

Technical Defensibility (30%)

• Is the agent architecture fundamentally different from a chatbot or automation tool?
• What data does the agent accumulate that makes it harder to replace?
• How does the agent improve over time as it learns your workflows?

Go-to-Market Clarity (20%)

• What's the initial wedge into the enterprise?
• Can they prove ROI with early customers?
• Is there a clear expansion path to adjacent workflows?

Market Timing (10%)

• Is the enterprise market ready to adopt agents in this workflow?
• Are there macro tailwinds (labor costs, competition, regulation) pushing adoption?
• Is there a window before incumbents can move?

Fenton would weight these differently than most investors. He'd put more emphasis on workflow understanding and less on the size of the TAM. He'd rather back a founder with deep domain expertise building a $50M market than a technical founder building a $5B market.

This is why his track record is so strong. He's not chasing the biggest markets; he's backing the founders who understand the workflow well enough to own it.

The Expansion Playbook

Once a founder has nailed the initial workflow, Fenton's playbook suggests a specific expansion path:

Phase 1: Own One Workflow

Start with sales qualification, customer support, or invoice processing. Get to 10 customers, prove 30%+ efficiency gains, establish clear ROI. This is the beachhead.

Phase 2: Expand Within the Customer

Take the agent into adjacent workflows within the same customer. A sales agent that owns qualification can expand to pipeline management, forecasting, and deal analysis. This is where you get land-and-expand economics.

Phase 3: Expand Across Departments

Once you own sales, move into marketing, customer success, and operations. Each department has similar workflows that need rebuilding. The agent infrastructure you built for sales scales across the organization.

Phase 4: Become Infrastructure

Eventually, the agent company becomes so embedded in the enterprise that it becomes infrastructure-like Zendesk did for support or New Relic did for monitoring. At this stage, the TAM expands dramatically because you're not just solving one workflow; you're the platform that orchestrates all agents.

Fenton would back a company at Phase 1 and expect it to reach Phase 4 within 10 years. That's the arc of his best bets.

What This Means for the Agent Wave

The agent wave is real. The opportunity is real. But the winners won't be determined by who has the best AI model. They'll be determined by who understands enterprise workflows deeply enough to rebuild them.

Fenton's thesis-identify where incumbents can't adapt, back the team that rebuilds the workflow-is more relevant now than ever. The incumbents (Salesforce, ServiceNow, Oracle) are trying to add agents to existing products. They can't rebuild because they're locked into legacy architecture and legacy customer expectations.

Meanwhile, new companies can be born with agent-first architecture. No technical debt. No installed base expecting the old workflow. No sales team trained to sell humans productivity tools.

This is the moment. And Fenton, based on his track record, will recognize it and back the right companies.

For founders building agent companies, the lesson is clear: Don't try to out-AI the incumbents. Out-understand them. Know the workflow better. Know the customer better. Know the pain points better. Rebuild the entire workflow, not just add an AI feature.

That's how you get Fenton's attention. That's how you build a $5B company.

For investors trying to understand where the agent wave creates value, the lesson is equally clear: Look for founders with deep domain expertise rebuilding workflows, not AI experts adding features. Look for companies that accumulate proprietary data and workflow context. Look for clear expansion paths from one workflow to many.

Fenton has been making these bets for 20 years. His track record suggests he knows what he's doing. And his thesis suggests the agent wave is just the beginning.

Key Takeaways for Founders Raising Capital

If you're a founder in the agent space, here's what you need to internalize:

1. Lead with workflow insight, not AI capability. Investors like Fenton care about whether you understand the enterprise workflow deeply enough to rebuild it. They don't care whether you're using GPT-4 or Claude.

2. Prove early unit economics. Get to 3-5 customers and show that your agent delivers measurable ROI. Fenton doesn't need a massive TAM; he needs evidence that the workflow you're rebuilding is valuable.

3. Articulate the expansion path. Show how you'll go from one workflow to many. This is where the real value creation happens, and it's where Fenton's valuations reflect the full potential.

4. Build defensibility through data and lock-in. The best agent companies will have high switching costs because the agent has learned your specific workflows and accumulated proprietary data about your business.

5. Understand your customer's constraints. Enterprise AI adoption is limited not by AI capability but by integration complexity, change management, and the need to prove ROI. Build your product and pitch around these constraints, not around the AI.

For more detailed guidance on pitching AI projects to investors, A Step-by-Step Guide for Entrepreneurs on How to Pitch Their AI Projects and Raise Private Money provides a structured framework.

Fenton's career suggests that the agent wave is just the latest iteration of a pattern he's recognized for two decades: when new technology emerges, the winners are the founders who understand the old workflow well enough to rebuild it for the new era. If you're building an agent company, that's the thesis you need to internalize.

The Future of Fenton's Agent Bets

Where will Fenton's agent thesis take him next? Based on his track record, we can expect:

More vertical agent companies. Fenton will back deep domain experts building agents for specific workflows in specific industries. These companies will have higher valuations and better defensibility than horizontal AI platforms.

Agent infrastructure plays. Fenton will back the companies that become the plumbing layer for agent orchestration. These companies will have similar characteristics to New Relic-essential infrastructure that every enterprise depends on.

Agent-native platforms. Eventually, Fenton will back companies that rebuild entire enterprise software categories from scratch with agents as the core abstraction. These will be the highest-risk, highest-reward bets.

The macro trend is clear: enterprises will move from human-centric workflows to agent-centric workflows over the next decade. Fenton's capital will follow that trend, backing the founders who understand the transition deeply enough to lead it.

For founders, the opportunity window is open. The capital is available. The market is ready. The question is whether you have the workflow understanding and the founder-investor fit to build one of the companies Fenton will back.

Based on his track record, the answer is yes-if you're the right founder with the right thesis about how the workflow will be rebuilt.