Miles Grimshaw on Venture Investing, Founder Partnerships and AI Business Models

In this September 2023 interview, Benchmark general partner Miles Grimshaw tells Harry Stebbings how he entered venture capital. Growing up, he watched his stepfather build an e-commerce software company at their kitchen table and became fascinated by people creating businesses.

As he finished university, New York’s early cloud and application ecosystem brought him into contact with Josh Kushner and Will Gaybrick as they built Thrive Capital. They invited him to join. He says the experience combined his interest in entrepreneurship with a rapidly developing technology scene.

Looking back, Grimshaw wishes he had more fully appreciated the scale of the cloud transition. He recalls comparing Shopify and Etsy and now thinks the important point was that both could benefit from a large shift online, even though their products and eventual outcomes differed.

His advice to new investors is to resist being overwhelmed by venture terminology and ask who the user is and why the product matters to them. He recommends working backward from the customer rather than beginning with abstract investing frameworks.

From Kushner he learned that kindness and competitiveness can coexist, and that believing in another person’s potential can be powerful. He also values determination and curiosity alongside demonstrated skill. His view that passion leads to extra work is a personal perspective, not a universal requirement.

A third lesson is being active and urgent in the present while remaining patient and maintaining standards over the long term. Stebbings says this tension is difficult when building an investment firm: progress matters each week, but a durable institution takes years.

Before joining Benchmark, Grimshaw says the most important consultation was with his wife about moving their family from the East Coast to California. At Benchmark he was encouraged to be himself rather than fit a prescribed role.

He describes Benchmark partnership meetings as free-flowing blocks of time, without a standard deal-flow sheet, formal presentation or agenda. Partners share what they have learned and discuss entrepreneurs and opportunities. He says trust in one another makes this style workable.

Grimshaw’s five parts of venture investing are sourcing, selecting, signing, supporting and “summiting”: ultimately distributing value to shareholders or fund investors. He emphasizes that an investment outcome is distinct from a company’s ongoing life.

The work he enjoys most is connecting strategy to sequencing, organization and execution. He wants diligence to feel like planning an expedition with a founder: serious questions about resources and people matter, but the aim is to look forward together rather than conduct an examination.

He encourages founders to meet investors before a pitch feels perfect. As an early-stage firm, he says, Benchmark can work through an idea with a founder even when the product and metrics are immature. This describes his firm’s approach and is also an invitation from an investor seeking opportunities.

He recalls a young technical founder who asked what customer lifetime value meant during a pitch. Grimshaw saw willingness to admit a gap and learn quickly as a strength. Stebbings disagrees, arguing that a software founder should know the fundamentals; Grimshaw says experience, context and openness to learning matter.

Asked whether venture investors add value, Grimshaw says a trusted partner can help a founder make better decisions over many years. He stresses that the founder makes the decisions. Trust, context and frequent communication are necessary for advice to be useful.

He warns investors to avoid harm: applying generic checklists without understanding the product and strategy can burden a founder. A partner does not need an answer to every question, but should be willing to work through the problem. He thinks great founders could do important work without a particular investor, while a good partner may help.

Grimshaw prefers thinking of an investment as a commitment rather than a bet. Stebbings notes that some founders see financing more transactionally. Both acknowledge that relationships vary, and Grimshaw says the best partnerships aspire to something deeper.

Rather than rank founder, market and traction separately, he considers how they interact. He describes himself as a “data-second” investor: begin with a theory about product and customer, then use data to validate it. He warns that early revenue growth can look impressive without building durable demand.

One test he uses is whether he would be excited to recruit people to the company. Hiring requires explaining the mission, team and opportunity repeatedly, so it forces an investor to assess the whole proposition. He also asks what has changed and why a founder is positioned to act on it.

He cautions against assuming the world will change simply because it ought to. A convincing investment thesis needs a strategy for adoption and an actual opening in the market. He tries to work backward from the user and look for evidence that the conditions for change are emerging.

At Benchmark, partners may meet a company together, although one typically takes the board seat. Grimshaw describes the firm as an equal partnership where founders can draw on other partners as well as their primary relationship. He says partners regularly help across one another’s investments.

On reserves, Grimshaw says Benchmark generally puts most of its intended capital into the initial partnership and keeps reserves for difficult circumstances, rather than planning to buy additional ownership later. This is the firm’s reported approach, not a general rule for fund construction.

He estimates that 50–60% of Benchmark investments at the time were the company’s first institutional partnership and around a third began at inception. These are his approximate firm figures. He says the absolute number of investments is small and that founders should not assume an idea is too early for the firm.

Grimshaw distinguishes realized successes, such as HubSpot and Segment, from still-unrealized holdings such as Benchling and Monzo at the time. He describes Benchling as a founder matched to a specialized market that expanded, while Monzo pursued a large, established banking market. His examples carry investment interests.

He praises Monzo founder Tom Blomfield’s commitment to the mission, including repeated crowdfunding rounds despite legal complexity. He cites account numbers as of the interview and acknowledges the company’s ultimate outcome was still ahead. Those figures and the investment value should not be read as current or realized returns.

His biggest missed opportunities include Figma, Plaid and Scale AI. He says he recognized strong founders but became too hesitant about their markets. These are retrospective assessments, not evidence that a different decision would necessarily have produced the same result for his fund.

With Figma, he underestimated how many people beyond designers would collaborate around a digital product: marketers, writers, product managers and engineers. He says user data showing non-design participation was available, but he did not give it enough weight. His later customer-spending observations are anecdotal.

With Plaid, he focused too much on how bank account authentication should work technically and too little on institutions’ incentives and users’ existing friction. He says the error was becoming overly intellectual rather than fully examining the system as it operated.

Grimshaw’s breadth across sectors comes from curiosity and from wanting to work with particular founders. He aims to become a specialist in each company’s situation rather than impose a fixed procedure. He says lessons from Stripe helped Segment, but treating Segment as a copy of Stripe could also be harmful.

He admits specialists can know metrics and benchmarks more deeply than he does, and tries to connect founders with that expertise. His own preference is to look for unusual adaptations rather than assume every company follows an established playbook.

In 2022, he became interested in language models as he saw examples of reasoning-like behavior and tool use. He expected AI development to spread beyond specialist machine-learning teams to many application developers. This was his thesis, not proof that every developer would become an AI engineer.

That thesis suggested a need for frameworks and tooling to build and evaluate applications using probabilistic models. It helped lead him to invest in LangChain and founder Harrison Chase. Benchmark has a financial interest in the company, and the prospective market remains uncertain.

Grimshaw says analogies between AI and mobile or the internet are imperfect. Mobile created new access and uses but, in his view, often retained existing business models. He thinks AI could alter application architecture and business models more substantially, while recognizing that this is a forward-looking thesis.

He summarizes the potential change as “selling the work, not the software.” Instead of only promising software availability, a provider might agree to deliver a business outcome, such as leads or completed accounting work. These are hypothetical models, not established capabilities or service guarantees.

He argues that a control center for a manager, with tools performing portions of the work, could differ from a per-seat assistant for an individual worker. He has seen little of this fully realized and says better models and time to build reliable products may be necessary.

He calls the copilot approach a natural incumbent strategy because existing vendors control distribution, user interfaces, data and pricing. Stebbings asks whether a startup could begin with a copilot and later deliver complete work. Grimshaw says a business model shapes product and distribution, making such a transition difficult but not impossible.

Stebbings worries that enterprises may adopt these systems slowly. Grimshaw replies that a startup only needs enough customers to grow initially, and a product that delivers a useful outcome at a lower price could be compelling. This is a scenario, not evidence of enterprise adoption or realized savings.

Both discuss possible employment effects. Grimshaw is optimistic that people will adapt and acknowledges potential needs around redistribution or taxation. His optimism does not establish how many jobs will be created or lost, or how quickly affected workers could adjust.

On media, Stebbings worries that AI-generated content will increase competition for attention. Grimshaw thinks trusted publishers may gain importance when text and images are easier to generate, and mentions The New York Times as an illustrative view. This is an opinion expressed in 2023, not investment advice.

Asked whether AI startups are merely model wrappers, he says some develop distinct technology while many current software features resemble copilots. He expects new products to take time because teams need capable models, fine-tuning and methods to manage probabilistic behavior.

Benchmark has invested in infrastructure including LangChain and Cerebras, he says. He believes tooling can benefit from broader AI development, while valuable applications could also emerge over time and pull infrastructure forward. These are investment theses from a firm with positions in the sector.

Rapid signups can reflect novelty rather than lasting demand, he warns. He says teams may reinvent and the final outcomes are uncertain; he does not claim all early AI investments will fail. Stebbings adds an observation about experimental corporate budgets and churn.

Grimshaw says he and LangChain’s founder spent roughly three months discussing the open-source project and its strategy before partnering. He did not want to pressure the founder into starting a company, because that decision carries responsibility. The company subsequently raised more money.

He says too much capital can encourage waste, but the outcome depends on people and culture. At LangChain, he says, hiring and planned initiatives rather than the availability of money should determine spending. This is his account of a portfolio company, not an independent financial assessment.

He reiterates that copilots suit incumbents but argues startups may build different architectures and business models. Stebbings suggests AI might also create new distribution channels; Grimshaw agrees it could open underserved segments while distinguishing that from the internet itself as a distribution channel.

The conversation turns to running. Grimshaw describes his own recreational routine and earlier endurance races, including a team event lasting about 60 hours. His food and training habits are personal anecdotes, not nutrition or exercise advice.

He says becoming a parent has made him more present and patient. Reflecting on business leadership after the low-interest-rate period, he praises the willingness to reverse a major decision when evidence changes, citing Shopify’s acquisition and subsequent divestment as an example.

He credits investor Michael Abramson with clarity of thought and praises Benchmark partner Eric for proactive, respectful board work. He favors “founder respect,” including candid disagreement, over mere cheerleading. The founder remains the decision-maker, he says.

Grimshaw says he made only one investment between late summer 2020 and the end of 2022 and does not regret it, while acknowledging he may have missed others. He finds the discipline of Benchmark’s early-stage focus challenging because curiosity draws him toward later-stage opportunities too.

Asked what LangChain could become in ten years, he describes a potential framework and tool ecosystem for developers building AI applications. The comparison with earlier developer platforms is his aspiration as an investor, not a prediction guaranteed to occur. The interview ends with mutual thanks.