Eiso Kant has worked on language models for code since the Sourced years and now co-leads Poolside. This profile examines his ideas about AI for software development and the practical work of engineering leadership. — Latent Space — Eiso Kant Interview.

Part 1: AI, AGI, and Poolside
- On code as a proving ground: Kant sees software development as a useful route toward broader AI capability because code follows defined rules and solutions can be checked more directly than many real-world tasks. — Air Street Interview.
- On feedback from code execution: Poolside’s training premise is to have a model generate possible solutions and reasoning steps, then use feedback from running and checking code to reinforce paths that produce correct results. — Air Street Interview.
- On agents in engineering workflows: At Poolside, Kant says agents increasingly write code, launch training jobs, evaluate results and modify research pipelines, while researchers remain responsible for directing the work. — Latent Space — Inside the Model Factory.
- On synthetic training data: Kant argues that coding tasks can generate additional training examples because multiple candidate solutions can be tried and checked against a verifiable result. — Air Street Interview.
- On the model factory: Kant describes model development as an engineering system that connects data, training, experiments and evaluation so researchers can move from an idea to a trustworthy result faster. — Latent Space — Inside the Model Factory.
- On screen-based and physical work: Kant expects work done behind a screen to be reshaped by AI sooner than some physical and relational roles; he contrasts near-term knowledge-work change with a more distant possibility of robotic surgery and the continued value of human care. — World Economic Forum — Eiso Kant Interview.
- On the AI capability gap: Kant views software engineering as a likely first major economically valuable field in which AI approaches human-level capability; he presents this as Poolside’s forecast, not an accomplished result. — World Economic Forum — Eiso Kant Interview.
- On headcount and output: Kant cautions that doubling an engineering team does not automatically double its software output; onboarding, dependencies and newly promoted leaders can slow delivery as an organization scales. — Refactoring — Scaling from 5 to 250 Engineers.
- On human context in an AI world: Kant suggests that people may retain value through information and relationships AI does not automatically possess, including the trust and context exchanged face to face. — World Economic Forum — Eiso Kant Interview.
- On research and engineering: Kant argues that building frontier models is heavily an engineering task: researchers write code and inspect data, while dedicated systems engineers make experiments reproducible and fast. — Latent Space — Inside the Model Factory.
Part 2: Engineering Metrics and DORA
- On the limits of DORA metrics: Kant argues that high-level delivery metrics are useful but too flat to diagnose a bottleneck on their own; leaders need to inspect the underlying pipeline and team context. — Hybrid Hacker — Eiso Kant on Metrics.
- On data as a conversation starter: Kant favors bringing metrics into engineering conversations, while warning that a metric can become the topic rather than a tool for understanding the work. — Developing Leadership — The Midwit Trap.
- On individual performance metrics: Kant cautions that tools centered on individual performance can feel like surveillance to engineers; Athenian instead focused on team workflows and events. — TechCrunch — Athenian Interview.
- On organizational visibility: As an engineering organization grows, leaders lose direct visibility into delivery details; Kant recommends looking for bottlenecks systematically and pushing more problem-solving responsibility to teams. — Refactoring — Scaling from 5 to 250 Engineers.
- On tracking outcomes: Kant recommends examining how engineering effort is allocated among features, bugs, technical debt, maintenance and research, then connecting that allocation to business goals and customer impact. — Refactoring — Scaling from 5 to 250 Engineers.
- On single-metric focus: Kant recommends starting with one leading metric, such as PR Cycle Time, and reviewing it consistently across the engineering organization to prompt useful conversations and change. — Hybrid Hacker — Eiso Kant on Metrics.
- On engineering leadership’s two responsibilities: Kant describes engineering leadership as improving the developer experience while delivering impact to end users; the two responsibilities reinforce each other through continuous improvement. — Scaleway CTO Forum — Eight Mental Models.
- On workflow visibility: Kant’s preferred analytics give teams visibility into planning, review and release stages so they can investigate delivery bottlenecks without ranking individual developers. — TechCrunch — Athenian Interview.
- On the goal of telemetry: Engineering metrics are useful when they tell leaders where to investigate a delivery problem and help teams improve developer experience and customer impact. — Refactoring — Scaling from 5 to 250 Engineers.
- On evaluating quality: Kant recommends quality measures close to the end-user experience, including bug severity, time to restore service, and change failure rate, rather than treating delivery speed alone as success. — Refactoring — Scaling from 5 to 250 Engineers.
Part 3: The Midwit Trap and Complexity
- On specialization and context: Kant values deep specialists, but argues that they need the self-awareness to know when their specialty is relevant and when a wider problem calls for another approach. — Developing Leadership — The Midwit Trap.
- On tool obsession: Kant warns that engineers can make a tool part of their identity and forget what problem it was meant to solve; he uses Docker’s advocacy program as one example. — Developing Leadership — The Midwit Trap.
Part 4: First Team Mode and Executive Alignment
- On first-team alignment: In his discussion with Jason Warner, Kant asks how managers at any level can move their teams toward company-wide alignment and argues for addressing tensions openly instead of optimizing only within a function. — Developing Leadership — Company Culture.
- On company-wide priorities: Kant urges leaders to surface difficult cross-team issues and optimize for what the company is trying to achieve, rather than only for a single team’s preferences. — Developing Leadership — Company Culture.
- On adaptability as teams grow: As an engineering organization grows, Kant argues that leaders must combine conviction with adaptability: building alignment becomes more important than simply winning an argument about a technical choice. — Scaleway CTO Forum — Eight Mental Models.
Part 5: Customer Discovery and Developer Experience
- On idea validation: Kant recommends using non-leading customer interviews to test an idea before committing to hires, spending, code, or a product. — Traction Interview.
- On finding an urgent need: Customer discovery should distinguish polite interest from an idea customers regard as urgent; Kant says that finding the latter earlier could save extensive wasted effort. — Traction Interview.
- On the SMB ladder: Kant advises learning from the smallest relevant customers first, then moving upmarket one or two rungs at a time as the product, sales knowledge, and credibility develop. — Traction Interview.
- On the developer-experience team: When an engineering organization reaches roughly 20–100 people, Kant recommends a team dedicated to developer experience to help manage growing dependencies and review bottlenecks, while keeping product teams accountable. — Refactoring — From 5 to 250 Engineers.
Part 6: Engineering Leadership and Expectations
- On respecting senior technical expertise: Kant reminds engineering leaders that a long-tenured staff engineer may know more about a system than their manager; leadership should make room for that expertise rather than assume managerial rank confers deeper technical context. — Scaleway CTO Forum — Eight Mental Models.
- On setting expectations for managers: Kant urges engineering leaders to tell newly promoted managers clearly what their role must achieve and communicate, instead of relying only on coaching or hints to convey expectations. — Scaleway CTO Forum — Eight Mental Models.
Part 7: Developer Productivity and PR Cycle Time
- On PR Cycle Time: Kant uses PR Cycle Time to examine the code-delivery pipeline, defining it as the elapsed time from a pull request’s first commit until its code is used in production. — Refactoring — Scaling from 5 to 250 Engineers.
- On detecting bottlenecks early: Kant recommends tracking lead time from work in progress to released pull request so a growing team can spot delays early, rather than relying only on the number of completed changes. — Refactoring — From 5 to 250 Engineers.
- On prompt review responses: Kant’s code-review advice follows Google’s original guidance: respond to a review request promptly so authors can keep moving, while treating first-response delay as a team-flow issue. — Google Engineering Practices — Review Speed.
- On review-stage bottlenecks: Kant warns that team dependencies can leave code reviews blocked as an organization grows; he recommends clearer processes and goals to help teams address that delivery friction. — Refactoring — From 5 to 250 Engineers.
- On PR size: Kant recommends keeping pull requests small and focused on one logical change, with enough clarity and context that a reviewer can understand them without a large interruption. — Eiso Kant — Smaller Pull Requests.
- On review speed and focus: Kant points to Google’s review guidance for a practical balance: respond quickly at a natural break, but do not interrupt concentrated coding work merely to start a review immediately. — Google Engineering Practices — Review Speed.
- On workflow friction: As a team scales, Kant recommends examining where dependencies and delivery bottlenecks appear, then setting explicit goals and giving teams responsibility for improving flow. — Refactoring — Scaling from 5 to 250 Engineers.
- On continuous improvement: For Kant, a metrics program is also change management: consistent attention to a leading measure can create the discussions needed to improve how an engineering organization works. — Hybrid Hacker — Eiso Kant on Metrics.
Part 8: Scaling Organizations and Making Bets
- On position sizing: In a discussion with Jason Warner, Kant treats engineering initiatives, technology choices and promotions as bets whose size should reflect available resources and the risks at stake. — Developing Leadership — Placing Bets.
- On reversibility and impact: Kant adapts a decision matrix by considering how reversible a choice is and how much impact it has, rather than sorting it only by urgency and importance. — Developing Leadership — Good Decisions.
- On scaling communication: As engineering teams grow, Kant warns that dependencies and communication lines increase; he recommends clearer processes and goal definition before the organization becomes much larger. — Refactoring — Scaling from 5 to 250 Engineers.
- On technical-debt trade-offs: Kant accepts some technical debt while an early team pursues product-market fit, but advises protecting the ability to scale and shifting more effort toward bugs and debt as the product and customer base mature. — Refactoring — Scaling from 5 to 250 Engineers.