20Product: Inside Legora's Tech Stack: Why Token Maxing is Failing...
The Twenty Minute VC (20VC)Full Title
20Product: Inside Legora's Tech Stack: Why Token Maxing is Failing Enterprise Startups with Jacob Lauritzen, CTO @ Legora
Summary
This episode features Jacob Lauritzen, CTO of Legora, discussing the evolving tech stack and engineering practices in the age of AI.
The conversation highlights how AI is reshaping software development, from code generation and review to product management and organizational efficiency, while also raising concerns about security and the changing nature of engineering roles.
Key Points
- The current software development process has shifted from code writing being the bottleneck to product ideation and code review becoming the primary areas for optimization, driven by AI tooling.
- AI is significantly increasing engineer productivity, making it cheaper and faster to write code, which necessitates a re-evaluation of bottlenecks and team structures.
- The role of engineers is evolving from pure code creation to higher-level system design, architecture, and enabling AI agents to optimize and self-improve systems.
- AI-driven code reviews are emerging but are still in their nascent phase, with a need for tools that focus on system architecture and design impacts rather than just lines of code.
- Prototyping with AI significantly speeds up the product development cycle, allowing product managers to front-load work and iterate without immediate engineering involvement, potentially shortening or altering traditional design phases.
- While AI can automate much of the functionality and design, "taste" remains a crucial differentiator, reflecting an opinionated stance and brand identity that prevents products from converging into a homogenous "grayness."
- AI-generated code is becoming prevalent, but concerns about new security threats are significant, necessitating continued human oversight in code reviews for enterprise applications.
- The efficiency gains from AI are impacting processes like postmortems, with agents now capable of quickly analyzing logs and metrics to help generate reports.
- Building internal tools with AI, or "vibe coding," is becoming increasingly viable and cost-effective for enterprises to customize systems beyond off-the-shelf solutions.
- The value of a product like Legora relies not just on the underlying AI models but on the surrounding primitives, enterprise features, and model optimization that create a cohesive and efficient user experience.
- The focus in AI model usage is shifting towards performance and latency, especially in domains like legal, where a slight delay for better output is acceptable, but real-time interaction is key.
- Open source models are crucial for sovereignty and security, and there's a growing need for robust European and American open-source AI models to avoid reliance on a few dominant players.
- The future of Integrated Development Environments (IDEs) will likely move away from focusing on lines of code to more graphical, architectural views where agents facilitate the implementation of designs.
- The cost of AI tooling is justifiable by its potential for significant efficiency gains and productivity increases, framed as an opportunity cost rather than a mere expense.
- Underestimating growth and delaying investment in developer experience and strategic hiring are common mistakes for rapidly scaling companies.
- Hiring for senior management roles in engineering is challenging due to the need for individuals who can balance technical expertise with leadership and manage scaling teams without compromising quality.
- Co-location and collaborative team structures are highly valued for reducing handover costs and increasing efficiency, even as remote work becomes more prevalent.
- The key to retaining top talent is focusing on challenging problems and strong team culture rather than solely on titles or compensation.
- The future of legal practice, much like coding, will likely involve working at a higher level of abstraction, focusing on strategy, risk assessment, and negotiation rather than minute textual details.
- Competing against large, established players requires persistent hard work and a willingness to outwork the "800-pound gorilla," as those within large organizations may not be as motivated.
Conclusion
The rapid evolution of AI is fundamentally changing the software development lifecycle, shifting bottlenecks from code writing to product ideation and review.
Companies must adapt by embracing AI tooling to increase engineer productivity, focusing on system design and architecture, and fostering a culture of continuous learning and reinvention.
The future of engineering involves a blend of human strategic oversight and AI-driven execution, with "taste" and unique problem-solving remaining key differentiators.
Discussion Topics
- How will the increasing reliance on AI for code generation and review fundamentally alter the skill sets required for engineers in the next five years?
- What ethical considerations and new security vulnerabilities arise as AI plays a larger role in enterprise software development and internal tool creation?
- As AI streamlines many aspects of product development, how can companies maintain a distinct brand identity and "taste" to avoid product homogenization in a competitive market?
Key Terms
- AI Tooling
- Software applications and platforms that leverage artificial intelligence to assist in various tasks, such as coding, debugging, design, and analysis.
- Bottleneck
- A point of congestion in a system that limits its overall throughput or efficiency.
- Code Review
- The systematic examination of source code by others to identify errors, improve quality, and ensure adherence to standards.
- DevOps
- A set of practices that combines software development (Dev) and IT operations (Ops) to shorten the systems development life cycle and provide continuous delivery with high software quality.
- Enterprise Startups
- Companies that aim to serve large organizations or businesses, often with complex needs and longer sales cycles.
- FDE (Field Engineer)
- A role often focused on customer implementation, support, and ensuring successful adoption of a company's product in the field.
- Inference Providers
- Services that offer the computational infrastructure and software to run AI models and generate predictions or outputs.
- Product Management
- The organizational process of strategy, creation, delivery, and management of a product or services throughout their life cycle.
- SRE (Site Reliability Engineering)
- A discipline that incorporates aspects of software engineering and applies them to infrastructure and operations problems.
- Token Maxing
- In the context of AI, this likely refers to maximizing the usage of AI tokens, potentially for performance benchmarks or to drive usage metrics, which may not always align with efficient or effective AI utilization.
- Vibe Coding
- A term used in the transcript to describe rapidly building custom internal tools or solutions using AI, often for specific business needs.
Timeline
The conversation begins by discussing how AI tooling has dramatically increased developer productivity, making code writing faster and shifting the focus to product work and code review as new bottlenecks.
The discussion delves into the changing role of engineers, moving from writing code to system design, architecture, and enabling AI agents.
The potential for AI code review to become a dominant force in the development lifecycle and its implications for reducing bottlenecks is explored.
The future of engineering is framed as a shift towards systems design and architecture, with AI handling code creation and maintenance.
The need for specialized teams to manage and optimize AI agents is highlighted, similar to developer experience teams.
The effectiveness of AI agents is discussed in relation to API quality and the importance of setting proper guardrails for their operation.
The percentage of AI-generated code versus human-generated code at Legora is discussed, along with concerns about future security threats.
The impact of AI on development processes beyond coding, such as PRs and postmortems, is examined, with postmortems becoming more efficient.
The role of Product Managers (PMs) in the AI era is debated, particularly how rapid AI-powered prototyping affects their workflow and the need for them to focus on product work rather than engineering.
The conversation addresses whether AI's efficiency in prototyping will lead to skipping the traditional design phase and the enduring importance of design language and taste.
The difficulty of copying products in the age of AI is discussed, emphasizing that the last 10% of a product's quality, dealing with edge cases and scale, is what truly differentiates it.
The differing speeds of AI development, product delivery, and human adoption are discussed in the context of bringing enterprise customers along.
The potential for "vibe coding" internal enterprise tools is explored, questioning whether this is the future or if Legora is at the forefront of innovation.
Specific examples of Legora "vibe coding" internal tools are provided, illustrating the practical application of AI in creating custom solutions.
The debate on whether to buy off-the-shelf HR systems or build custom ones using AI is analyzed based on system complexity and customization needs.
The evolving role of Product Managers in the next few years is discussed, focusing on the trade-offs between PMs doing engineering and the opportunity cost of neglecting core product work.
Advice is given to aspiring engineers on the importance of continuous learning and adaptability in a rapidly changing technological landscape.
The extent to which Legora's product quality depends on underlying AI models is discussed, highlighting the importance of surrounding features and infrastructure.
The future of open-source AI models is considered, emphasizing their growing role and the need for diverse geographic contributions.
Concerns about the lack of European and American open-source AI models are voiced, highlighting potential geopolitical and competitive risks.
The potential for a new, non-code-focused role in enterprises related to internal AI systems management is predicted.
The necessity of Field Engineers (FDEs) for driving enterprise adoption of AI tools is discussed, with the expectation that this may decrease over time.
The hardest roles to hire for today are discussed, with senior management and engineering directors being particularly challenging.
The discussion highlights the importance of a developer experience team to facilitate fast onboarding and efficiency for engineers.
Differences in hiring engineers in Europe versus the US are explored, focusing on risk aversion and employee loyalty.
The discussion touches on the ecosystem creation within companies and the problem of "token maxing" by encouraging excessive AI token usage.
The role of tools like Cursor in optimizing token spend for enterprises is highlighted, with a debate on the impact of recent acquisitions.
The potential death of traditional IDEs and the emergence of new forms of development interfaces are contemplated.
The willingness to invest in AI tooling for developers is framed around opportunity cost, emphasizing the high value placed on efficiency gains.
Mistakes made in the past, such as underestimating growth and delaying investment in developer experience, are reflected upon.
The impact of scaling for 100x usage versus 10x on system design and the importance of considering burstiness and fair queuing are discussed.
The choice between having superior AI models or superior engineers for a six-month advantage is presented, with a preference for engineers.
The most significant learning from the early days of Legora is the underestimation of the company's rapid scaling.
The concept of founders being destined for specific company stages is dismissed in favor of problem-solving capability and adaptability.
The secret to hiring engineers with no ego is discussed, emphasizing that true talent prioritizes challenging problems over titles.
The importance of co-location for efficiency and reducing handover costs in product development teams is highlighted.
The preference for engineers who desire collaborative problem-solving over purely remote work is stated, aligning with Legora's culture.
Predictions for Legora's engineer headcount by the end of 2027 are made, emphasizing a preference for quality over quantity.
The speed of acquiring top talent through company acquisitions versus direct hiring is debated.
The integration process for acquired companies is described as surprisingly easy when hiring low-ego engineers focused on problem-solving.
Personal reflection on past hiring mistakes stems from a lack of experience in leading engineering teams and underestimating senior talent.
The process of providing strong feedback to new hires and the short timeframe for determining a successful fit are detailed.
The hardest roles to hire for are identified as senior management and engineering directors.
The company's stance on managers is clarified: they must be highly technical and capable of full-stack work, with accountability for team health.
The involvement of the CTO in new product features is explained, with a focus on significant launches and strategic product vision.
Max's role as a salesman and product visionary is highlighted, emphasizing his belief and conviction.
The impact of a particular Sifted article on Legora and the company's reaction are discussed.
The success of a "Jude Law" brand campaign is discussed, highlighting its viral nature and ability to generate conversation.
A quick-fire round begins with questions on changed opinions and underrated AI companies.
The biggest threat to Legora is identified as the inability to continuously reinvent itself.
Potential sports team sponsorships for Legora are discussed, with F1 and the Yankees being mentioned.
A crazy prediction about the future of law, drawing parallels to the evolution of coding, is shared.
Advice to founders competing against "800-pound gorillas" is to simply work harder.
Episode Details
- Podcast
- The Twenty Minute VC (20VC)
- Episode
- 20Product: Inside Legora's Tech Stack: Why Token Maxing is Failing Enterprise Startups with Jacob Lauritzen, CTO @ Legora
- Official Link
- https://www.thetwentyminutevc.com/
- Published
- June 6, 2026