Head of AI

What matters most

• Tie AI strategy to firm outcomes

  The role sets direction with leaders across investing, portfolio impact, and operations, then owns a focused roadmap and impact metrics. Prepare to explain how you would prioritize opportunities by their potential to improve sourcing, diligence, deal wins, or portfolio support.

• Build a trustworthy shared context layer

  Primary wants firm knowledge captured systematically and made safely accessible to AI through reusable infrastructure and evaluations. Prepare to discuss how you would assess context quality, access boundaries, and product reliability.

• Turn real workflows into adopted products

  The posting calls for observing how investors and operators work, finding the problem beneath a request, prototyping quickly, and shipping products that improve decisions. Prepare an approach to workflow discovery and evidence of adoption, without assuming a particular use case is already planned.

• Make hands-on technical trade-offs

  The Head of AI must evaluate LLMs, agents, retrieval, context layers, and evaluation approaches, while contributing directly to builds when needed. Prepare to explain how you decide whether to prototype, build, buy, or stop.

• Lead a coordinated firm-wide effort

  The role directs engineers, analysts, and other contributors; enables teams to build agents on shared foundations; and clarifies ownership across functions. Prepare to describe how you would communicate decisions and keep distributed work aligned without duplicating it.

Likely challenges to think through

• Capture useful signal without overexposing it

  A likely challenge, not a confirmed problem: valuable relationship knowledge may be fragmented or sensitive, while AI products need usable context. Prepare to reason through the trade-off between broad retrieval and safe access, including how you would test whether captured context is accurate and useful.

• Measure impact beyond activity

  A likely challenge: shipped tools and usage are easier to count than improvements in investment or portfolio outcomes. Prepare to distinguish adoption and decision-quality indicators from longer-term firm results, and explain how you would revise a roadmap when evidence is weak.

• Balance local experimentation with shared standards

  A likely challenge: letting teams closest to the work build agents can generate useful ideas but also duplication and inconsistent reliability. Prepare an approach to shared foundations, evaluation, and ownership that preserves room for fast experiments.

• Earn adoption while changing established workflows

  A likely challenge: an AI product may be technically strong yet fail to fit how investors or operators make decisions. Prepare to explain how you would validate a workflow problem before building, seek feedback on prototypes, and decide when to change or stop a product.

Topics to prepare

• AI strategy tied to venture outcomes

  The posting calls for an AI plan across investing, portfolio impact, and operations, measured by better deals sourced, won, and supported. Suggested revision exercise: compare two possible AI initiatives and explain which firm outcome each targets, what evidence would justify prioritizing it, and what would cause you to stop.

• Shared context and safe access

  Primary wants a shared context and data layer that captures knowledge from relationships and workflows and makes it safely accessible to AI. Suggested revision exercise: diagram how insights from a customer call and a candidate reference could be captured, retrieved, and restricted to appropriate users.

• LLMs, agents, and retrieval tradeoffs

  The role explicitly requires current, hands-on fluency with LLMs, agents, context and retrieval layers, and the model landscape. Suggested revision exercise: compare a retrieval-based assistant with an agent for one diligence workflow; explain their differing complexity, failure modes, and conditions for use.

• Evaluation and trustworthy reuse

  The posting asks for evaluation methods and common technical foundations that make AI products trustworthy and reusable. Suggested revision exercise: define a small test set for answers grounded in firm knowledge, then specify how you would check answer quality, access boundaries, and regressions.

• Zero-to-one product judgment

  Primary seeks products that emerge from actual investor and operator behavior, earn adoption, and create measurable value. Suggested revision exercise: take an ambiguous request such as 'improve diligence' and distinguish the underlying user problem, a first prototype, an adoption signal, and a build, buy, or stop decision.

Skills to revise

• Turn workflow observation into a prototype

  The posting expects rapid movement from observing work to shipping useful products, with hands-on contribution when needed. Suggested practice: use fictional, non-sensitive call notes to sketch a diligence workflow and prototype a narrowly scoped way to surface relevant prior insights; identify where a human must review the output.

• Prioritize and measure an AI roadmap

  The Head of AI owns roadmap priorities, visible progress, and impact metrics. Suggested practice: draft a one-page roadmap for two hypothetical Primary workflows, stating the sequencing rationale, a near-term adoption measure, and a longer-term firm-outcome measure for each.

• Run a comparative AI evaluation

  The role involves assessing approaches and establishing evaluation methods. Suggested practice: create representative diligence questions and expected evidence, compare two LLM or retrieval approaches against them, and record quality, failure cases, and the tradeoff behind a recommendation.

• Design reusable agent foundations

  Primary wants tools, standards, and support that let teams build agents on shared foundations rather than duplicate work. Suggested practice: outline the shared context access, evaluation checks, ownership, and documentation that a team-built agent would need before broader firm use.

• Communicate decisions and lead across teams

  The posting calls for directing contributors, aligning senior stakeholders without formal authority, and explaining risks and results to technical and non-technical audiences. Suggested practice: write a brief decision memo on one proposed AI product, then rehearse a partner-facing summary and an engineer-facing explanation of its tradeoffs and ownership.

• Select a zero-to-one product case study

  The posting asks for adopted products with measurable value. Prepare a walkthrough of one relevant project, covering the initial ambiguity, user observations, your building and leadership contributions, adoption evidence, and measured results.

• Sketch an investor or operator workflow

  Primary names sourcing, diligence, winning deals, and portfolio support as areas AI touches. As an interview exercise, choose one workflow and outline the user problem, a prototype, how you would test it with users, and when you would build, buy, or stop.

• Draft a shared-context design

  The role leads a context and data layer that captures firm knowledge and makes it safely accessible to AI. Prepare a high-level design for capturing information from relationships and workflows, retrieving it with appropriate access controls, and reusing the foundation across products; label assumptions for discussion.

• Prepare a technical tradeoff brief

  The posting expects hands-on fluency with LLMs, agents, retrieval, and evaluations. Bring a concise example comparing possible approaches for your proposed workflow, including failure modes, an evaluation plan, and what you would prototype before committing.

• Build a sample priorities-and-metrics plan

  The Head of AI owns a roadmap tied to firm outcomes. Draft a hypothetical first set of priorities across investing, portfolio impact, and operations, with reasons for sequencing and proposed measures of adoption and decision impact—not claims about Primary’s current systems.

• Gather leadership and alignment evidence

  Prepare specific examples of directing contributors, aligning senior stakeholders without formal authority, clarifying ownership, and communicating technical risks to non-technical audiences. Identify your decisions and their observed outcomes in each example.

• Write questions for firm leadership

  Ask what AI work is already underway, where relationship knowledge is currently lost, how access and ownership are handled, which outcomes matter most, and how engineers, analysts, investors, and operators collaborate. Use the answers to test the assumptions in your proposed designs.