Guy Podjarny on AI Coding Tools and the Future of Software Development

Guy Podjarny, founder of Snyk and CEO of Tessl, joins Harry Stebbings to discuss AI models, developer tools and the future of software creation. Podjarny leads an AI development company and invests in startups, giving him a direct commercial interest in several topics they discuss.

Asked whether Nvidia is undervalued, Podjarny separates three questions: growth of the AI semiconductor market, Nvidia's share of that market and the price of its stock. He expects demand and competition to grow and sees durable advantages in Nvidia's manufacturing, intellectual property and CUDA ecosystem, but is less certain about valuation relative to other investments.

Stebbings asks whether disappointing returns on AI spending could reduce chip demand. Podjarny expects a period of enterprise disillusionment because budgets and expectations have moved faster than organizational processes. He believes the technology will keep improving, but predicts many projects will fail or duplicate competitors' work.

On the potential cost of AGI, Podjarny expects substantial investment but cannot validate a specific multitrillion-dollar figure. He says adoption also depends on law, insurance, accountability and social acceptance; even a capable AI lawyer or self-driving vehicle must fit institutions and rules.

He notes that AGI is poorly defined, making claims about its arrival difficult to evaluate. Stebbings suggests that fundraising incentives may shape optimistic timelines; Podjarny agrees incentives can bias expectations without necessarily implying bad faith.

He thinks training competitive general foundation models requires large amounts of capital. Specialized models for code or robotics might gain ground if larger models encounter limits, though he expects capital and general models to remain important over a longer horizon. He identifies reports of frontier-model difficulties as rumors, not confirmed facts.

Tessl decided not to train its own foundation model. Podjarny wants to use whichever external models best serve particular tasks, such as reasoning, coding or visual work, rather than compete with or commit to one model provider.

Despite criticism of coding assistants, he says many developers find them useful and do not want to return to their old workflow. He also worries that generated code can be mediocre or insufficiently reviewed: accepting a suggestion may involve less thought than writing and examining it oneself.

He attributes Cursor's appeal partly to low-friction adoption. It fits an existing development workflow, suggests changes across files and allows users to inspect them. A tool does not have to work perfectly every time if checking its output is easy enough.

Current tools can reduce repetitive work such as drafting documentation and tests, and code completion provides a useful starting point. Beyond these uses, Podjarny says reliability remains uneven: systems can be impressive on one task and fail unexpectedly on another, a pattern his colleague calls AI's 'jagged edge.'

He rejects the idea that generating similar code automatically recreates an established SaaS business. A company may have customer relationships, data, distribution, switching costs and ongoing service. If the only advantage is code that nobody else can write, he thinks that advantage may be fragile.

Podjarny defines agentic development by the amount of work delegated. At one extreme, a user requests an application and the system researches, designs, builds and verifies it. Most contemporary tools instead assume a developer remains involved and provide assistance with particular steps.

He describes a spectrum between small app-generation demonstrations, workflows in which humans and machines define software together, and more autonomous outsourcing. Tessl is pursuing a workflow-oriented approach; his description of competitors' internal methods is tentative.

On criticism of Devin, Podjarny says an impressive demo is easier to create than a reliably working product and believes early marketing overstated capabilities. He says people he has spoken with found Devin interesting but unreliable; that is anecdotal feedback rather than a controlled evaluation.

Stebbings questions large early financing rounds. Podjarny says model-training firms may need substantial GPU budgets, while Tessl spends on evaluation and iteration without training a foundation model. He believes reserves support a long effort but acknowledges that raising a lot reduces a company's future options.

He worries that closed development environments could concentrate software creation in a few powerful platforms, leaving other tools unable to connect or modify the process. Stebbings notes that foundation-model providers also want external developers to build on them; Podjarny sees the risk particularly in closed application-building layers.

Podjarny considers concentration a meaningful possibility, especially if platforms handle everything from code generation to deployment and developers surrender control. He treats this as a risk and prediction, not a description of a settled market.

He thinks some assistants already deliver value beyond experimental budgets. More autonomous systems can be useful in specific settings such as support, while sales automation can damage a brand when it goes wrong. He expects wider outcome-oriented adoption to take longer because processes and reliability must improve.

Podjarny predicts model providers may move deeply into applications using the same data, such as search or code generation. Specialized data and complex workflows could leave more room for independent companies. This is his view of a changing competitive landscape and reflects Tessl's position in it.

Given hypothetical valuations for OpenAI, Anthropic and xAI, he chooses Anthropic, citing the team and price. He says he does not know what is happening inside OpenAI but thinks substantial leadership turnover can disrupt an organization. His choice is an opinion under hypothetical terms, not investment advice.

On search, Stebbings argues Perplexity is much better than Google and that users can switch quickly. Podjarny responds that established habits and Google's distribution are powerful; he thinks Google can copy useful features even if its product execution is imperfect. They disagree on the speed of user change.

He predicts a change in US merger enforcement could benefit some firms more than the largest technology companies. On Snyk, he would like it to go public eventually for employee liquidity, transparency and brand recognition, but says the timing depends on costs and circumstances, not a confirmed plan.

Podjarny says handing over Snyk's CEO role was difficult even though he values the successor and the benefits of the transition. He returned to a CEO role at Tessl because he missed the job.

He expects the best developers increasingly to act as systems thinkers: understanding requirements, anticipating trade-offs and making architectural decisions. He predicts direct coding will remain but become a smaller part of the role over roughly ten years, especially for specialized or older technologies.

An architect must decide, for example, how much extensibility is worth added complexity or whether to optimize for one cloud versus portability. These choices depend on business priorities and future uncertainty; Podjarny does not assume a model can simply know the right answer.

Stebbings questions whether there will be enough architecture decisions for everyone. Podjarny expects more software to be created and sees a second path toward product work and user empathy. He predicts higher developer leverage and rising customer expectations rather than a fixed amount of software work.

Some technical product-manager roles may merge with development, he says, while people who understand users and make accountable decisions remain important. AI may assist their work. He does not know exactly what a technology company will look like in five years, but expects strategy, system ownership and management to remain.

Security becomes harder when generated code is poorly reviewed or has no clear owner. Podjarny warns that code can persist and become vulnerable as dependencies and systems change. He calls for maintenance, ownership and checks throughout creation, not merely a one-time generation step.

In a question round, Podjarny says he sold roughly one third of his Snyk ownership for a nine-figure sum. He recalls early acquisition discussions around $200 million and later, less concrete multibillion-dollar conversations. Those discussions were not completed transactions.

He recounts a fundraising meeting in which two investors appeared disengaged on their phones, leaving him feeling insulted. This is his account of the meeting, not an independent determination of their intentions.

Podjarny says he has revised his expectations about the arrival of next-generation large models and thinks architectural difficulties could slow progress for a year or two. He labels this a sense of the situation and a possibility, not a verified development schedule.

He wants founders to think from first principles about workflows that might no longer be needed, rather than only optimize today's tasks. His contrarian advice is to look for a valuable idea that requires convincing people: a proposal everyone immediately understands may attract many competing companies.

He favors markets with a handful of competitors over markets with none or dozens. Too few can indicate a lack of demand; too many make it hard to stand out. In crowded markets, he says operational and go-to-market execution matter as much as product quality.

Tessl had raised $125 million at the time of the interview. Podjarny says the cash provides flexibility and time, while interest income depends on the balance and rates. He does not present the discussion's rough $5–10 million interest range as guaranteed income.

He identifies two opposite errors after raising too much too early: spending rapidly before product-market fit and losing urgency because the runway feels endless. A larger team is harder to redirect, while moving too slowly can miss a market window. He says urgency should come from market timing as well as cash runway.

Among his angel investments, he says a Cloudinary position rose from almost nothing to about $4 million in value. He also mentions other investments, including a small early check in SecurityScorecard. The interview does not establish that the Cloudinary value was realized in cash.

Podjarny says money gives him freedom rather than happiness by itself. He and his wife are thinking about how to give away wealth; on marriage, he emphasizes communicating about problems instead of letting them quietly accumulate.

He admires Intercom's decision to build the autonomous support agent Fin. He also praises Lightdash's embedded reporting approach and discloses that he is an investor; his enthusiasm for its sales and future is a portfolio investor's account.

Stebbings says he may invest more in Lightdash, underscoring that both speakers have commercial interests in the example. They close by thanking one another for the discussion.