Overview of Inside Legora's Tech Stack: Why Token Maxing is Failing Enterprise Startups with Jacob Lauritzen, CTO @ Legora
Harry Stebbings speaks with Jacob Lauritzen, CTO of Legora (the fast-growing AI legal enterprise), about how AI is reshaping software engineering, product development, hiring, security, and enterprise workflows. The conversation centers on a core thesis: the bottleneck has shifted from writing code to product thinking, review, systems design, and operational guardrails. Jacob argues that companies should not optimize for “token maxing” or AI usage vanity metrics, but for real output, speed, and business impact.
Main Themes and Takeaways
1) AI has made coding cheap — the new bottlenecks are elsewhere
- Legora’s team uses tools like Claude Code, Cursor, and other AI assistants to ship, debug, and iterate faster.
- Jacob’s view: the hardest part of software used to be writing code; now that’s become much cheaper.
- The new constraints are:
- code review
- product discovery / scoping
- system design
- security and governance
- He expects engineers to move “one layer up” from code writing to systems architecture and strategic trade-offs.
2) Code review will be transformed by AI, but human oversight still matters
- Legora already uses AI code review, but Jacob считает it still early and not enough.
- He thinks the future of review is not checking every line manually, but reviewing:
- architecture impact
- security boundaries
- stability risks
- whether the change pushes the system in the right direction
- He wants a world where AI handles most implementation, while humans focus on high-level judgment.
3) Product management becomes more important, not less
- In Jacob’s view, AI makes it easier for PMs to prototype quickly and validate ideas earlier.
- That means PMs can do more:
- talk to customers
- synthesize feedback
- form opinions on product direction
- test prototypes before engineering gets involved
- But he does not think PMs should become full-time coders.
- Reason: the real bottleneck is now product work, so if PMs spend too much time engineering, the company loses valuable discovery and synthesis capacity.
4) “Taste” and opinionated product direction still matter
- Jacob strongly believes AI will make it easier for products to converge visually and functionally.
- That makes taste more important as a differentiator:
- having a clear stance
- knowing what the product is and is not
- resisting generic AI-generated sameness
- He sees taste as an organizational identity, not just aesthetics.
5) Legora is building internal AI systems, not just external products
- He predicts a major enterprise role will emerge: internal AI systems / AI enablement.
- This team would build and govern the internal tools, workflows, and guardrails that let agents operate safely inside an enterprise.
- Examples include vibe-coded internal tools for:
- HR systems
- talent acquisition
- payroll
- employee onboarding
- migration support
- Legora already uses this approach internally to build custom tools cheaply and quickly.
Engineering, Security, and AI Model Strategy
6) AI-generated code is already a major share of output
- Jacob says Legora’s top code producers are AI tools like Claude and Cursor.
- Human and AI-generated output are now close enough that AI is a major contributor to the codebase.
- This is why the company is investing heavily in:
- developer experience
- guardrails
- review systems
- onboarding automation
7) Security risk is rising alongside AI-generated code
- Jacob is very concerned about AI-accelerated threat actors and vulnerabilities.
- Legora still reviews all human PRs because enterprise security matters.
- He expects future systems to use risk scoring and more automated review, but believes current defense tooling is not yet good enough.
8) Model quality matters, but not as much as product systems around it
- Legora frequently evaluates models from OpenAI, Anthropic, and open-source options.
- Jacob says model choice changes often and depends on the task.
- Key selection criteria:
- performance first
- latency second
- cost later
- He argues customers do not buy Legora just because of the underlying model; they buy the surrounding product, legal primitives, and enterprise reliability.
9) Open source models are strategically important
- Jacob is bullish on open source because it supports:
- sovereignty
- security
- competition
- He worries about a future dominated by too few model providers.
- He wants stronger European and American open-source model ecosystems.
Hiring, Org Design, and Culture
10) Legora underestimated hiring needs
- Jacob admits one of his biggest mistakes was thinking the team would stay tiny.
- He once believed the company would cap out at around 20 engineers; they are now around 80 and still scaling.
- He now plans for 100x scale, not 10x scale.
11) Developer experience became a critical function
- He says Legora should have invested earlier in a developer experience team.
- That team now builds:
- faster local dev environments
- custom background agents
- review automation
- onboarding tools
- He thinks better dev tooling can materially raise engineer productivity.
12) Hiring philosophy: low ego, high trust, strong technical depth
- Jacob emphasizes hiring people who:
- have no ego
- care more about solving problems than titles
- can work well in a small, intense team
- He’s comfortable giving very direct feedback quickly.
- He believes integrating small, high-quality teams is easier than trying to manage big hierarchical orgs.
13) The company is highly in-person and collaborative
- Being together in Stockholm matters a lot to Legora’s speed.
- Jacob believes co-location reduces handoff costs between PMs, designers, and engineers.
- He’s opinionated against overly siloed, remote-first workflows for the kind of fast-moving, collaborative work Legora is doing.
Future Predictions and Advice
14) IDEs and coding workflows will change dramatically
- Jacob thinks the current shape of the IDE will disappear.
- He expects future software development to focus less on lines of code and more on:
- system visualization
- architecture planning
- agent orchestration
- He even suspects the next-generation “IDE” may be graphical rather than text-centric.
15) Don’t optimize for token usage — optimize for outcomes
- His advice to enterprises: don’t create token-maxing incentives.
- Reward:
- output
- speed
- efficiency
- business impact
- He says spending on AI tooling should be judged by opportunity cost:
- if AI saves time and creates advantage, spend aggressively
- if not, don’t overuse it just to look impressive
16) For founders facing a giant competitor: outwork them
- His blunt advice for startups competing with an “800-pound gorilla”:
- move faster
- stay lean
- work harder
- He argues large incumbents often have complacency and less urgency than small teams.
Notable Quotes / Soundbites
- “The bottleneck is no longer coding.”
- “Taste is what separates us.”
- “Don’t token max. Optimize for output.”
- “If you’re a big enterprise, you need guardrails so agents can run inside the system safely.”
- “The thing that can kill us is if we stop reinventing ourselves.”
Practical Takeaways
- Build workflows around speed and iteration, not just raw code generation.
- Invest early in:
- developer experience
- review automation
- security guardrails
- AI-enabled internal tools
- Keep product teams focused on customer insight and synthesis.
- Use AI to reduce handoff friction, not to replace all human judgment.
- For enterprise AI, the winning stack is not just the model — it’s the product layer, workflows, controls, and trust.