# 20Product: Inside Legora's Tech Stack: Why Token Maxing is Failing Enterprise Startups with Jacob Lauritzen, CTO @ Legora

Podcast: The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
Published: Jun 17, 2026
Reading time: 21 min
Canonical: https://podbrew.app/briefs/the-twenty-minute-vc-20vc-venture-capital-startup-funding-the-pitch-20product-in

This edition of Podbrew podcast notes features Jacob Lauritzen, CTO at Legora, a B2B enterprise company recognized for its rapid growth, achieving $100 million ARR in just 18 months. Jacob shares deep insights into Legora's tech stack and the challenges facing enterprise startups today, particularly around the concept of 'token maxing'.

The discussion explores the evolving landscape of software development, where AI tools are redefining developer productivity and shifting bottlenecks from code creation to product definition and code review. Jacob highlights the changing role of engineers towards system architecture and strategic trade-offs, alongside the emerging need for 'meta-engineering' to optimize AI agents. He also addresses the critical security implications of AI-generated code and the imperative for human review.

Lauritzen provides essential perspectives on designing for extreme scalability, scaling large product teams, and Legora's unique approach to talent acquisition and developer experience. The notes underline why continuous reinvention, an ego-less culture, and outworking competitors are crucial for hyper-growth companies in the current technological climate. These insights offer valuable lessons for any organization navigating rapid growth and the transformative impact of artificial intelligence.

## Key takeaways

- AI tools like Cloud Code and Cursor have drastically increased developer productivity, making code writing a much less time-consuming and expensive part of software development.

- The main bottleneck in software creation has shifted from the act of writing code to optimizing code review processes and refining product definition.

- AI code review is becoming dominant, with bots handling much of the iterative review process, thereby removing it as a human bottleneck.

- Engineers' responsibilities are shifting from coding and code review to higher-level concerns like system architecture, design, stability, and strategic trade-offs.

- A new role, "meta-engineering," involves optimizing AI agents, setting guardrails, and defining custom rules to ensure their effective and safe operation within a system.

- AI-generated code from tools like Claude and Cursor now accounts for a significant portion of new code creation, introducing novel security vulnerabilities.

- Human review of all pull requests remains critical for enterprises to ensure security, as current automated defenses are inadequate against AI-introduced flaws.

- AI enables Product Managers to rapidly prototype, test, and iterate on ideas independently, streamlining early development and validating concepts before engineering resources are committed.

- While AI can accelerate the development of product functionality, the broader aspects of design
—such as establishing a consistent design language, user experience, and brand "taste"
—remain essential for product differentiation and identity.

- "Taste" is a crucial differentiator in the AI era, representing a company's unique, opinionated stance that prevents products from becoming generic and ensures they convey a distinct identity to users.

- With AI making coding cheaper, the new bottleneck shifts to strategic product work, including customer research and synthesis.

- Legora employs "vibe coding" to rapidly develop custom internal tools, finding it more efficient to build bespoke systems like talent acquisition and payroll from the ground up rather than heavily modifying existing third-party software.

- Legora actively manages a diverse portfolio of about ten AI models, continuously evaluating and switching between them bi-weekly based on task-specific performance and latency, not cost, to ensure optimal output.

- A dedicated developer experience team is crucial for optimizing local development environments and building custom tools that significantly enhance engineering efficiency and workflow automation.

- Utilizing AI tooling, like cloud coders and AI-powered documentation, streamlines the onboarding process for new engineers by providing instant answers to their questions.

- An intelligent AI usage strategy should reward efficiency and output, not just token usage, to avoid 'token maxing' and ensure genuine productivity.

- Traditional IDEs will evolve significantly, moving from line-by-line coding to a planning and agent-driven code generation model.

- Enterprises should view AI tooling investment for developers as near-infinite due to the high opportunity cost of not maximizing efficiency in a competitive market.

- All new systems are designed for 100x scalability to handle rapid, unforeseen usage spikes, especially for bursty operations like bulk data extraction.

- Effective leaders should continuously evaluate their role and be willing to yield their position if someone else can solve current problems faster.

## 04:00 - 08:01 AI's Impact on Engineering Org Design and Productivity

Building an engineering organization today, particularly at a large company like Legora, requires a fresh approach due to the rapid evolution of development practices. Jacob Lauritzen notes that his lack of prior experience in large orgs is an advantage, allowing him to adapt without preconceived notions from older methodologies.

A key driver of this change is the widespread adoption of AI tooling, such as Cloud Code and Cursor. These tools have dramatically increased developer productivity, allowing engineers to produce significantly more code than before and consequently shifting how teams are structured and operate.

Historically, the primary bottleneck in software development was the speed at which engineers could write code. With AI making code generation much cheaper and faster, this bottleneck has largely been removed. The focus has now shifted to the two other critical phases: efficient code review and the initial product definition work.

The challenge for modern engineering leaders is to optimize these new bottlenecks. This includes streamlining review processes and enhancing the efficiency of product managers in translating user needs, strategic priorities, and internal vision into tangible, iterative product work.

> The rate limiter was how quickly can you write code. That is now super cheap, so that's sort of been compressed. And so the, the bottleneck now is like the two other ends, which is review, how can we do that much more efficiently? And then it's how can we actually do the product piece much more efficiently?

## 08:00 - 12:01 AI Code Review Shifts Engineering Focus to System Architecture

AI is rapidly becoming the primary method for code review, with specialized bots and agents handling the iterative process. These AI reviewers can even engage in a sort of "fighting" pattern until a solution is reached. This automation is poised to eliminate the bottleneck of human review for individual lines of code.

The shift means human engineers are freed from low-level code inspection. Their attention is redirected to high-impact areas such as systems architecture, overall design, stability, and security boundaries. Engineers will focus on strategic trade-offs and ensuring the system evolves in the right direction, rather than reviewing every code change.

Consequently, the engineer's role is evolving from direct code creation and maintenance to a higher-level abstraction. They will increasingly think about the overarching system design, identifying opportunities for reuse and ensuring overall system stability. This changes the core job description from writing code to designing the ecosystem where AI agents operate.

A new discipline, "meta-engineering," is emerging, dedicated to making AI agents highly effective. Similar to developer experience teams, these engineers will focus on optimizing agent performance, setting up essential guardrails, and implementing custom rules to guide agent behavior. This role involves enabling agents to self-improve and independently optimize system objectives like conversion rates.

> The job of an engineer is changing from typing a bunch of code to sort of one layer above it, which is what does the system look like?

## 12:01 - 14:01 AI-Generated Code Creates Significant New Security Risks

A growing percentage of new code, led by tools like Claude and Cursor each contributing around two percent, is now generated by AI, surpassing the output of any single human engineer. This rapid adoption introduces a new generation of security threats by creating vulnerabilities that were previously unknown.

Despite the push for efficiency, ongoing human review of pull requests remains essential to mitigate these new risks. Companies like Legora still conduct thorough human reviews, acknowledging that current automated defenses are insufficient to manage the complexity and scale of potential AI-introduced flaws.

The concern is that AI-generated code allows for a much larger number of varied attack vectors, requiring equally robust and rapid defense mechanisms that are not yet fully developed. This gap contributes to an observed increase in security incidents and hacks, as companies struggle to adapt their defenses.

The speed at which AI can generate and potentially exploit code means that current security postures are often outpaced, leading to more frequent breaches. This necessitates a fundamental reevaluation of security processes and tools to keep pace with AI's capabilities.

> The next generation of security threat with the amount of AI-generated code that, bluntly, opens vulnerabilities we didn't know we had.

## 14:01 - 18:02 AI Accelerates Prototyping and Reshapes the Role of Design

AI is significantly accelerating the software development lifecycle, particularly by empowering Product Managers to rapidly prototype ideas. PMs can now quickly develop, test, and iterate on concepts themselves, delaying the need to involve engineers until a prototype clearly demonstrates value. This shift allows for much faster initial validation of new features and products.

While AI streamlines the development of functionality, potentially allowing companies to skip traditional detailed design phases for individual features, the role of broader design remains critical. Design still holds a vital place in establishing a consistent design language, overall user experience, and the "taste" of a product. Tools like Figma continue to be essential for maintaining this higher-level design consistency across larger systems.

"Taste" emerges as a key differentiator in a world where AI can quickly generate functional prototypes. It represents a company's unique, opinionated stance and identity, which prevents products from becoming generic or "grainless" as AI-driven development becomes more prevalent. This distinct taste allows a product to stand out and communicate "who we are and what we do," even if it means not appealing to everyone.

The ultimate goal remains to build valuable solutions for clients as efficiently as possible, but not at the expense of quality. Despite rapid prototyping capabilities, thoroughness in handling edge cases and unhappy paths is crucial to avoid merely "vibe coding" and to ensure a robust, reliable product.

> You need to have taste to have sort of an opinion and a stance in the world.

## 18:02 - 22:03 Bridging the Gap Between AI, Product, and Human Adoption Speeds

Jacob Lauritzen highlights a fundamental challenge in product development: the speed of AI and product creation often far outpaces the speed of human adoption. Legora focuses on translating the immense speed of AI development to historically underserved user bases, aiming to improve their efficiency and productivity by removing burdensome tasks and allowing them to focus on more strategic work.

Internally, Legora extensively uses what they call "vibe coding" for rapid tool development. This approach involves building custom internal systems from scratch, such as talent acquisition or payroll, rather than attempting to heavily customize existing off-the-shelf solutions that often don't fully meet specific needs.

This method proves efficient and cost-effective for creating bespoke tools. For example, an employee vibe coded an interactive app in just one day to guide Canadian team members through the migration process to Sweden, detailing laws and steps and saving significant time for the entire team.

Vibe coding enables the creation of both large, complex systems and numerous smaller, highly specific applications. This strategic internal development approach helps Legora optimize operations and customize tools precisely to their requirements, ultimately enhancing overall company efficiency.

> you can build product so much faster than your customer can consume it.

## 22:03 - 24:03 The product manager's role shifts as coding becomes less of a bottleneck in the AI era.

Jacob Lauritzen suggests that the widely discussed convergence of product and engineering roles may not apply to all companies, especially those with a strong need for dedicated Product Managers. While some organizations might see a single person handle both product and building, this model can be inefficient where deep product strategy is crucial.

In the age of AI, coding is becoming increasingly efficient and less of a constraint. This shifts the primary bottleneck from engineering execution to strategic product work, such as understanding customer needs and synthesizing insights. Therefore, the core value of a PM lies in these higher-level product activities.

If Product Managers spend a significant portion of their time coding, it represents a high opportunity cost. This diverts them from the essential "product work" that is now the critical bottleneck. While some "vibe coding" for high-fidelity prototypes can be beneficial, extensive engineering work by PMs means missing out on vital customer-centric strategy.

> Focus on the bottleneck, and the bottleneck's no longer coding, which means the bottleneck is the product work.

## 24:03 - 26:03 Career Resilience and Advanced AI Model Strategy

Jacob Lauritzen advises new computer science graduates to prioritize continuous reinvention and learning. He stresses the importance of staying adaptable and at the forefront of technological changes, as new developments emerge weekly. This constant improvement is vital for long-term career resilience.

Legora's product value extends significantly beyond the underlying AI models. While models are necessary for its existence, the core value lies in legal-specific primitives, enterprise features, and an optimal routing system that intelligently selects the best model for a given task. Customers choose Legora for its comprehensive solution, not just the raw model power.

Legora employs a dynamic and sophisticated AI model strategy, frequently evaluating and switching between providers like OpenAI and Anthropic, sometimes as often as bi-weekly. They leverage approximately ten different models concurrently. For each task, Legora assesses models based on performance and latency, prioritizing these factors over cost to ensure optimal output quality.

> The most important thing is you need to figure out how you constantly reinvent yourself and keep learning and improving, because things change all the time right now.

## 26:03 - 28:03 Open-Source AI's Momentum and Geopolitical Concerns

Open source AI is currently experiencing a strong moment, with a wealth of excellent models that are easy to deploy and supported by efficient inference providers. This progress even enables local device applications, such as offline transcription or coding assistance on a personal computer.

The existence of robust open-source models is considered vital for reasons of national sovereignty and security, providing independent capabilities for nations.

A significant concern is the observed lack of European and American contributions to open-source AI models. This absence is problematic because a duopoly or monopoly on AI models would create an unfavorable situation from a game-theoretic perspective, potentially undermining competition and national interests. Europe, in particular, should play a role but has yet to establish one.

Without broader participation, especially from key regions, the future landscape of AI models could be dominated by a few players, which raises security and competitive concerns.

> I think it's an important thing that we have open source model for sovereignty reasons and for like security reasons, we should have great open source models.

## 28:03 - 30:03 Jacob Lauritzen Predicts New Enterprise AI Roles and Discusses Underestimating Scaling Needs

Jacob Lauritzen anticipates a significant shift in enterprise IT, envisioning a new

internal AI systems role,

> I consistently underestimated how many people we need to be.

## 30:03 - 32:03 Legora's Developer Experience Team Boosts Engineering Ramp-Up and Efficiency

Legora maintains an engineering team of approximately 80 people and emphasizes highly selective hiring. Despite being a relatively small team, the company prioritizes rapid ramp-up for new engineers to counteract potential slowness and enable building more features.

To achieve fast ramp-up and enhance overall developer productivity, Legora has invested in a dedicated developer experience (DX) team. This team is responsible for optimizing the local development setup, ensuring it is fast, efficient, and quick to spin up.

The DX team builds custom tooling, including a background coding agent that allows engineers to run up to ten concurrent local development environments. They also develop review agents that automate continuous integration processes, ensuring all checks and reviews are green before involving a human for final approval.

Additionally, the DX team creates onboarding tools and emphasizes comprehensive README files in repositories. They leverage AI tooling, such as a "cloud coder" or "cursor," to enable new engineers to quickly find answers to questions, significantly accelerating their integration and effectiveness.

> Make sure you have really good README files in your repository so that a new engineer will just ask their cloud coder, their cursor, about all their questions. Like that's remarkably effective.

## 32:03 - 34:04 Contrasting European and US Engineer Hiring with Smart AI Tooling

Hiring engineers in Europe differs significantly from the US due to cultural attitudes towards risk and loyalty. European candidates are typically more risk-averse and exhibit greater loyalty once committed to a company, although it may take more time to convince them to join. In contrast, the US hiring market is described as more transactional, with individuals being less risk-averse and more willing to explore new opportunities.

A key difference lies in the understanding and valuation of equity. In Europe, especially in countries like Sweden, companies often need to educate potential hires about how equity works within the venture ecosystem. Unlike the US, where equity is more commonly understood as a future asset, Europeans are less familiar with its value and mechanisms, requiring detailed explanations of its non-cash benefits and tax implications.

Regarding AI tooling strategy, a common pitfall is 'token maxing,' where employees consume large amounts of AI tokens simply to inflate their usage metrics. This approach, driven by leaderboards or performance reviews focused solely on token consumption, is counterproductive. Instead, the advice is to foster intelligent AI usage by rewarding actual output and efficiency.

Effective AI integration should encourage employees to demonstrate how efficiently they leverage AI tools through initiatives like hack days or demos. The focus should be on celebrating tangible results and improved productivity rather than just the volume of tokens used, ensuring AI tools genuinely contribute to business goals.

> That's a really stupid way to do anything.

## 34:03 - 36:04 Traditional IDEs are Dying, AI Investment Should Be Near-Infinite

Jacob Lauritzen predicts the demise of traditional Integrated Development Environments (IDEs). He envisions a future where engineers primarily plan and review, while AI agents are responsible for generating the actual code. The focus will shift from meticulously reading lines of code to orchestrating and validating what the agents produce.

Lauritzen argues for a near-infinite investment in AI tooling for developers. He frames this as an issue of opportunity cost in a competitive landscape. The cost of not adopting these tools and gaining the significant efficiency improvements they offer far outweighs the investment, making it a critical strategic imperative for enterprises.

He also discusses the potential for tools like Cursor to optimize token spend for enterprises by routing requests to cheaper models or setting usage limits. However, Lauritzen expresses concern about vertical integration in the industry, specifically mentioning Cursor's acquisition by Grok, and suggests that model-independent platforms like Cognition and Factory might fare better by offering flexibility across various AI models.

> I think the current shape of an IDE will die, yes. I don't know what the new, the next IDE is, but it's not reading lines of code.

## 36:04 - 38:04 Scaling Systems for 100x Growth and Prioritizing Developer Experience

Jacob reflects on past missteps, specifically not adequately investing in developer experience and underestimating their product's growth. To address these issues, he has now mandated that all new systems must be built to scale for 100x their current usage, abandoning the previous 10x target which proved insufficient.

Designing for 100x scalability means carefully considering system limits, particularly in scenarios with 'burstiness'. For example, a tabular view product that allows bulk extraction from documents presents a significant challenge when scaling from 10,000 to 100,000 extracted cells due to the immediate spike in system load.

To manage these high-load spikes and maintain a good experience for all users, a fair queuing system is implemented. This approach means that users performing very large tasks, like extracting 100,000 cells, can expect to wait longer, while those with smaller, quicker tasks, such as extracting 10 cells, will experience much faster processing.

When faced with a choice between gaining access to a superior model or superior engineers for six months, Jacob emphatically chooses engineers. He explains that models are constantly evolving and improving on their own, but skilled engineers are the critical factor for achieving sustained, long-term exponential improvement within the company.

> Engineers, for sure? 'Cause the models, they change all the time, they get better all the time,

## 38:04 - 40:04 Legora's CTO explains his leadership philosophy of ego-less culture, continuous self-evaluation, and hiring for problem-solving.

Jacob Lauritzen reflects on Legora's rapid scaling, admitting he initially underestimated the pace and had to quickly adapt his leadership approach. He views his role not by title, but through the lens of problem-solving efficiency, continuously questioning if he is the fastest person to address the company's current challenges.

This ethos contributes to an ego-less culture at Legora, where individual titles are not paramount. Lauritzen openly states that if a team member demonstrates greater capability for his current role, he would be willing to swap positions or transition to another area. He consistently evaluates his own job performance and strives to rectify any deficiencies promptly.

When hiring engineers, Legora seeks candidates who prioritize tackling difficult problems over specific job titles. Lauritzen notes that top talent often values higher compensation more than a fancy designation. The interview process focuses on discussing the complex challenges applicants will work on, rather than emphasizing hierarchical roles. Being co-located in Stockholm is also crucial for minimizing handover costs and enabling rapid, collaborative problem-solving.

> Am I the person that can solve the problems that we have right now the fastest, or can someone else solve them faster than me?

## 40:04 - 42:04 Legora prioritizes co-location for efficiency and plans to grow its engineering team to 270 while retaining top talent.

Legora maintains a strong opinion that co-location significantly boosts efficiency by minimizing handovers and communication inefficiencies often seen in remote work, such as unclear Zoom calls or poorly written documentation requiring multiple reviews. They prefer engineers who want to solve problems collaboratively, rather than those seeking isolated remote tasks.

The company projects substantial growth, aiming to increase its engineering team from the current 80 employees to 270 by the end of 2027. This expansion requires adding approximately 190 new engineers within two years.

A key challenge with such rapid growth is retaining

A-level

> As soon as you have people that you don't trust or that the team doesn't trust, the A players won't stick around.

## 42:04 - 44:04 Using M&A to acquire talent and overcoming hiring judgment errors

Acquiring entire teams through mergers and acquisitions offers a significantly faster route to securing A-level talent compared to traditional individual hiring processes. Instead of recruiting two engineers per week, M&A can bring in a group of five highly skilled professionals at once. This strategy is particularly effective when targeting small startups whose teams are accustomed to collaborating closely.

When a startup team is acquired, their existing codebases are typically shed, and their knowledge is integrated into the acquiring company's platform, such as Lagora. The new hires become fully embedded members, bringing their expertise to the new codebase. Integrating these teams proves surprisingly easy if the individuals possess low ego and are primarily driven by a desire to solve problems rather than focusing on titles or organizational hierarchy.

Jacob admits to past hiring mistakes, often stemming from his own lack of experience running large engineering teams. He sometimes doubts his intuition, especially when interacting with seemingly more senior candidates who present confidently. Despite an initial sense that something might be amiss, he has occasionally convinced himself that the more experienced person must be correct.

However, these doubts often prove accurate within two to six weeks. Jacob notes that a hiring mistake typically becomes clear within a month, prompting him to provide very strong feedback to the individual after just two weeks to address performance issues.

> It's surprisingly easy if you hire people with low ego.

## 44:04 - 46:04 Legora's model for technical engineering managers and Max's focus on product vision

Hiring senior technical management is one of Legora's most difficult challenges. The company insists that its engineering managers remain highly technical and actively involved in coding. They explicitly reject a hierarchy of managers managing other managers who are not hands-on, advocating that non-technical managers should not be in engineering leadership roles.

Despite their technical focus, Legora's engineering managers are also crucial for team health and accountability. They lead small teams, typically six people with a PM, and are responsible for ensuring team members are performing well and are doing good. This blend of deep technical skill and team oversight ensures both productivity and well-being.

Regarding product development, Max, a key leader, focuses his time on high-level product vision and significant new features. For smaller features, the company empowers its capable senior engineers and self-organizing teams to make decisions and execute. This approach leverages the expertise of the team while allowing Max to concentrate on strategic direction.

> If you can't do full stack, get out, pick up your severance and go away.

## 46:04 - 48:05 Legora's viral marketing strategy, hiring insights, and biggest threat

The "Bloodsmack" campaign at Legora became an internal meme and a highly visible external marketing effort. The goal of the campaign was to generate widespread conversation, which it successfully achieved by being seen and discussed widely, even leading to internal screensavers.

Jacob also shared a significant change in his perspective over the last year, now believing in the importance of hiring more extensively. He operates under the principle that if adding a new team member results in a net positive for the company, then hiring should be pursued.

When asked about Legora's biggest threat, Jacob identified it not as external competition or specific individuals, but rather the company's own failure to continually reinvent itself. He stressed that stagnation would ultimately be the downfall of the company.

> The thing that's gonna kill us is if we don't keep reinventing ourselves.

## 50:04 - 50:05 Startups Can Outwork Giants to Compete Effectively

For startups competing against large, established companies, a key strategy is to simply "work harder." While bigger incumbents might boast greater resources and market share, a lean startup team possesses distinct advantages: passion, intense focus, and inherent agility. These qualities allow smaller teams to execute faster and dedicate more concentrated effort towards their objectives.

This dedication to outworking competitors can be a powerful differentiator. It helps overcome the natural inertia and bureaucracy often found in larger organizations, enabling startups to punch above their weight. By channeling their energy and drive, early-stage companies can achieve ambitious targets, such as significant revenue growth, effectively challenging the status quo through sheer determination.

> just work harder

---

Get podcast briefs for shows you follow: https://podbrew.app/