Jerry Murdock on Autonomous Agents, Software Investing and Adapting to Change

Harry Stebbings asks Jerry Murdock, a co-founder of Insight Partners, why he compares the coming AI transition to a tsunami. Murdock says the important wave is autonomous agents rather than AI as an undifferentiated category. His analogy is a warning that change may arrive in several stages and that companies should prepare before the full effects are visible.

Murdock does not predict that every software company will disappear. He argues that merely adding an AI feature may not be enough for a durable business, although it could help a company reach an exit. His preference is for teams that rethink their products around agents and follow the open-source communities building them.

He describes several AI-native startups in his portfolio and says they are experimenting with OpenClaw, NanoClaw or internally built agents to write code. The examples reflect his own investment relationships and the teams he has observed over a short period. They are not a representative survey of software development.

Murdock says some of those founders consider Cursor’s then-current product obsolete, but he personally believes its team, customers and capital give it a chance to adapt. The criticism is attributed to the founders he cites, while the prediction about Cursor’s response is his own. Neither establishes the future of the company.

He sees open-source communities as a source of integrations and experimentation that could expand agent capabilities. Comparing the emerging ecosystem to the Linux, Apache, MySQL and PHP stack that supported web development, he imagines an agent stack with tools, models and infrastructure that can work together. The analogy describes a possible path, not an established standard.

Murdock predicts an orchestration layer will route different parts of a workflow to different models: a more expensive model where quality is needed and an open model where it is sufficient. He expects agents eventually to test alternatives and select resources. Those capabilities and economic outcomes remain projections; the interview does not establish a universally optimal routing system.

He further predicts that more open models and specialized application-specific integrated circuits, or ASICs, will change the balance of AI infrastructure. He thinks specialized chips may be cheaper or better suited to particular workloads than general-purpose accelerators. The comparison depends on workload and implementation, and his account is a technology forecast rather than a verified cost benchmark.

Stebbings discloses a substantial Nvidia holding and asks how a shift to ASICs might affect it. Murdock speculates about Nvidia’s relationship with Groq and the role of CUDA in supporting future hardware. His characterization of the arrangement as an acquisition and his explanation of company motives are not established by the interview; he says the outcome would depend on execution.

Stebbings asks whether model routing will commoditize models and whether model providers will absorb application companies. Murdock thinks agents may increasingly compare alternatives by running workloads in separate environments. He does not provide a definitive answer about which companies will capture the value; his proposed autonomous selection is a prediction.

When Stebbings asks how to invest safely in applications that could compete with a model provider’s next release, Murdock says there is no safe answer. He estimates that around 80% of his own investments returned less than 1.3 times capital and says a smaller group generated much of his success. Those figures are his retrospective account, not audited fund-wide results.

On the fall in some public software and security shares after AI announcements, Murdock proposes that cautious buyers may be waiting for more information rather than all investors panicking. He describes this as his interpretation of market behavior, not evidence about the motives of individual traders or the future performance of named stocks.

He recalls the technology downturn beginning in 2000 and the difficulties that followed. Although he says Insight had avoided some dot-com investments, the broader software portfolio still suffered. The historical account is personal recollection; he does not claim the next cycle must reproduce the same declines or duration.

Murdock points to E2B, a company in which he has an interest, as an example of infrastructure designed for agent workloads. He reports a roughly 80-millisecond sandbox response versus an illustrative 400-millisecond human-facing threshold, and argues that agents may be sensitive to small delays at scale. These are his company and performance claims, not independently verified benchmarks.

Asked whether agents destroy or strengthen systems of record, Murdock says execution matters. He discloses a long-standing investment in Carta and offers a hypothetical: if tokenized securities use its existing records, the company could benefit; if a new system bypasses it, the existing position could weaken. Tokenization and either outcome are scenarios, not established future results.

He applies a similar question to Salesforce and the companies built on its platform. Murdock expects the large existing business to endure for some time, but says its value will depend partly on how its ecosystem responds to change. That is a framework for analysis, not proof that any particular platform or partner will succeed.

Murdock argues that market prices can react before the underlying business changes. In his view, existing software companies may use their customer context to become more valuable with agents, while others could decline if management adapts too slowly. The contrast is his forecast and does not establish the trajectory of a specific company.

Stebbings asks whether extraordinary growth rates have changed software expectations. Murdock imagines autonomous agents receiving identities and credentials, purchasing software within delegated limits and reporting their spending and results for human review. He presents this as a future operating model; the conversation does not establish that existing agents can safely or legally make every such decision.

He predicts that consumption-based pricing will fit some agent purchases better than seat-based pricing. Murdock mentions Docker as one of his companies and says it has moved toward usage-based pricing as it develops AI-related products. His view reflects an investment interest and an anticipated shift, not proof that one pricing model will suit every customer.

Stebbings asks whether incumbents can successfully bolt AI onto existing software. Murdock says trying is not the same as executing at a world-class level. He believes software that is useful to autonomous agents can thrive, while predicting challenges for products designed only around human buyers; the timeline and degree of adoption are uncertain.

The interview turns to employment. Murdock predicts that AI and labor displacement could become significant issues in a future US election, and that administrative, marketing and junior development hiring could be affected. His suggested timing and claims about which jobs agents can perform are forecasts, not measured unemployment or proof that whole occupations are replaceable.

He thinks small businesses may adopt agents faster than large enterprises because a single support or administrative role can be a large cost for a small team. He mentions a company using agents to help triage support emails. This example does not establish that its customer service requires no human oversight or that its experience applies to every business.

Murdock speculates that a minimum guaranteed income or expanded retraining program could become a political response to displacement. He describes a possible monthly payment and new work opportunities, but provides no enacted policy, financing plan or evidence that such a program will appear on a ballot. His account is a proposal and political prediction.

The guests consider whether smaller companies staffed by a mix of people and agents could improve quality of life. Murdock says the outcome would depend on what the remaining workers do and the culture they build; he does not claim to know the internal results at the company Stebbings cites. He also believes a billion-dollar one-person company is possible, a speculative future scenario.

Asked how private-equity-owned software might fare, Murdock recalls past buyout firms that struggled after technology and market shocks. He expects some investors to make poor decisions and others to adapt, without claiming knowledge of the financial condition or future of any specific named firm. His lesson is to revisit assumptions when conditions change.

Murdock says he is now more interested in creative impact than repeating his earlier career. If starting an investment operation today, he would use agents to gather and analyze data about markets and evaluate how founders use the tools. He presents this as a strategy he would consider, not a demonstrated advantage or a substitute for diligence.

Stebbings asks whether wealth made him a better investor. Murdock treats financial success as an imperfect proxy for experience and validation, and says his own decisions combine logic with intuition. He believes agents will lack intuition for some time; that is his view of future systems, not a proven technical limit.

Asked when intuition failed him, Murdock distinguishes intuition from wishful thinking. He says he sometimes backed founders he liked personally without adequately assessing their drive. His conclusion is a personal investing lesson, not a reliable rule that socially uncomfortable people are better founders or that likable founders underperform.

He discusses Fei-Fei Li’s work on visual representations and argues that language alone can be an incomplete description of the world. He expects richer representations to help future agents. That is his interpretation of a research direction, not proof that any system has an objective understanding of reality.

Stebbings asks about the possibility of regretting a different career. Murdock recalls a conversation with actor Kenneth Branagh and describes choosing to remain based in Aspen while co-founding Insight in New York. His advice is to consider whether a chosen path supports personal growth rather than assume a different one would necessarily have been better.

Murdock tells younger partners that wealth by itself does not provide self-understanding and that mistakes can be a source of wisdom. Stebbings shares his own experience with alcohol and failure. The exchange is personal reflection, not a suggestion that harmful experiences are necessary for growth.

Murdock recalls traveling to Kenya as a young student and spending time documenting traditional dances with local communities. He describes the unfamiliar experience as formative. These are his memories of that period, including his own account of the project, rather than a general account of the communities involved.

Looking back at Insight, he says an early attempt to establish a separate European operation was a mistake: the firm was not mature enough and needed its team close together to learn and make decisions. He distinguishes that organizational choice from the European investments the firm had already made.

On remote management and investment decisions, Murdock says effectiveness depends on the team and experience, though he found repeated travel and remote decisions difficult himself. He is uncertain whether a future mix of people and agents changes the trade-off. His preference is grounded in his experience, not a controlled comparison of working arrangements.

Asked about timing and a large Insight fund, Murdock emphasizes fund vintage as an important influence on opportunities and outcomes. He contrasts vintages positioned before the mobile wave with later ones, using examples from memory. His timing thesis is an opinion and does not establish the performance of the particular fund mentioned by Stebbings.

He calls the current transition a potentially attractive time to start a fund because new investors can build around agents from the beginning. That assessment depends on his view of adoption and does not guarantee investment returns. He says he retired from Insight’s routine negotiations, board seats and committee work but has continued investing with people he trusts.

Murdock recalls Insight’s 2009 investment in Twitter, when he says the business had a small team and no revenue. He saw the product as a status update for the world and credits his partners for supporting the decision. His assessment of later management conflicts is retrospective and he says he does not know whether a particular founder dismissal was right or wrong.

He says losses have been a recurring part of his investing career and that learning to endure them helped him maintain confidence. In the quick-fire round, he suggests graduates become familiar with agent tools and praises Insight’s sourcing partners. His advice reflects his beliefs and professional relationships, not a guarantee of employment or investment success.

He remembers meeting Truecaller’s founders in Aspen and eventually investing despite initial hesitation, after they traveled to see him. He says the company later went public. The story illustrates the importance he places on personal relationships and is his recollection of that investment.

Asked to choose among AI companies at hypothetical valuations, Murdock favors OpenAI, citing a reported 800 million users. He acknowledges Google’s distribution advantages and says execution and the ability to convert users to agent products matter. The user figure and valuations are unverified, and his selection is a personal investment opinion.

On money, Murdock says it comes without instructions and should be treated as a resource that can help people build useful things. He reflects on becoming a parent later in life, saying he does not know whether his earlier self could have balanced children with building Insight. These are personal values and counterfactual reflections.

He identifies surviving the post-2000 downturn as a formative moment for Insight and describes staying together when many firms struggled. Asked about parenting, he says children observe their parents and encourages being present and acting with care. His account of the firm and family advice are personal recollections.

The conversation closes with hopes that AI might improve longevity. Murdock repeats a claim that surviving five more years could add another five years to life, and Stebbings also anticipates major gains. These are speculative hopes and an unverified claim, not established medical evidence, a prognosis or a guarantee of life extension.