Build civilization-scale infrastructure.

Apik does not run a single recruiting funnel against a single role-shape. The archetypes below are deliberately distinct: the work each of them does is different, the signal we look for in each is different, and the reasons we decline a strong candidate from one archetype are not the reasons we would decline from another. The shapes are the ones that have stabilized across the frontier-research cohort, and the descriptions here are written so a candidate can self-assign before applying.

Hiring at Apik is slow on purpose. A typical loop runs four to eight weeks from first contact to offer; senior hires can run longer. The slowness is the signal — if the loop is moving fast, the loop is calibrated against the wrong standard. We benchmark the bar against the public hiring documentation and the published practice of the frontier-research cohort: the calibration target is Anthropic-tier and DeepMind-tier hiring, not generic-AI-startup hiring.

We do not run leetcode loops for research-track roles. The signal we have found most predictive — across both research-scientist and research-engineer archetypes — is the ability to write a two-page argument that survives substantive review. Strong candidates can produce an argument-bearing document about a contested research question, and their argument holds up against the obvious counterarguments. The signal is unusually decision-relevant because the company itself runs on documents of this shape: research agendas, policy frameworks, system cards, manifesto-style framings. A candidate who cannot produce documents like this would not have the leverage at Apik that the role requires.

The standard interview shape: an initial conversation about research interests and prior work; a written exercise on a question relevant to the role; a technical round with a researcher in an adjacent pillar; a culture and posture round with one of the founders; references. We share interview material in advance where possible, and we publish feedback to declined candidates where they ask for it. The bar is high; the process is meant to be respectful regardless of outcome.

Apik publishes compensation bands for every role at offer time. The bands are calibrated against Anthropic-tier and DeepMind-tier base-cash for equivalent roles; equity is meaningful and is structured to reward sustained tenure rather than departure. Where a role's band is wider than usual — typical of research-scientist seniority bands — the position within the band is decided by calibration against current Apik teammates at the same seniority, and the calibration is explained at offer time.

We publish bands because the compensation literature is unambiguous about what unpublished bands do: they widen the asymmetry of negotiation in favor of the employer, in ways that compound across cohorts and disproportionately disadvantage candidates from groups under-represented in the field. The transparency posture is the same one that runs through the rest of the lab — the Responsible Development Policy, the system-card schema, the policy review on news posts — and it is meant to be of a piece.

We do not run counter-offer dynamics. We do not refresh equity to retain against an external offer. We do not pay above-band to win a signed candidate from a competing offer. The principle is that the band is the band, and the band has to be defensible at the start of the loop or it is not the right band.

The visiting-researcher program runs three-to-twelve-month engagements for senior researchers and PhDs who have a defined collaboration in mind. The program is funded; the deliverable is a publishable artifact — a paper, a policy framework, a verified subsystem, a demonstrated component — that would not have existed without the collaboration. The program structure is modeled on Anthropic's residency framing, MIRI's research workshops, and the visiting-researcher norms at academic-affiliated centers like Cambridge's CSER, Oxford's GovAI, and the OpenAI Forecasting program.

The scientific case for visiting researchers is direct: rotating outside expertise prevents the inbreeding that a small, focused lab is otherwise structurally prone to. Specific pillars — interpretability, formal methods, mechanism design, the embodiment substrate — are research areas where a three-to-six-month deep collaboration with an external specialist materially changes the outcomes the lab produces. The Responsible Development Policy and the system-card schema both benefit from external review by visiting policy researchers and visiting safety scientists; that review has been built into the document-review processes.

Visiting researchers receive Apik-internal access proportionate to the collaboration scope, attribute their work to themselves as primary author with Apik as the institutional collaborator, and retain copyright on the artifact unless the collaboration is structured otherwise. Proposals go to careers@apiksystems.com with a one-page brief, the proposed deliverable, and at least one reference who can speak to the candidate's prior work.

Apik is remote-first with two on-site weeks per year (one research retreat, one engineering retreat). Most communication is written and asynchronous; the standing meetings are kept few. Slack is for coordination, not for work; the work happens in documents, code, and review threads. We do not run a standup; we do run a weekly research review where anyone — researcher, engineer, visiting fellow — can present current work and have it critiqued by the room. The review is the canonical site of internal feedback and is intentionally substantive.

Reading-group cadence runs at one paper per week per pillar, plus a cross-pillar reading group on the manifesto-relevant literature. Attendance is voluntary; participation in the discussion is expected for full-time research roles. The reading list is curated by pillar leads and is published internally; the reading-group discussions occasionally convert into news posts on this site.

Deep work is the default expected mode. Calendars are unscheduled by default for research-track roles; meetings are scheduled into the week with preference for batching. The norm is that two consecutive uninterrupted days per week are the floor, not the ceiling. The tradeoff is fewer status meetings and slower-feeling coordination; the gain is the kind of sustained engagement that is the operating premise of the lab.

For research-track roles: a cover letter that makes a specific argument about what you would push on at Apik, two to three pieces of relevant prior work (papers, repositories, posts), and references who can speak to recent work. We do not gate on CV format; a clean academic CV or resume is fine but is not the signal we read against. Cover letters under one page that argue something contestable beat cover letters under three pages that argue nothing.

For research-engineer roles: a cover letter, a representative code artifact (production code is fine; open-source contributions are fine), and a short description of one engineering decision you made on a recent project that you would defend or revisit. The engineering loop is calibrated to assess judgement under realistic constraints, not algorithmic agility under artificial ones.

For visiting-researcher proposals: a one-page brief on the collaboration, the proposed deliverable, the specific Apik person or pillar you would work with, and a senior reference. Proposals that are pillar-aligned and have a named Apik counterpart in mind are read first.

All applications go to careers@apiksystems.com. We respond to every applicant within two weeks of receipt; declined candidates can request feedback and we will provide it where the loop generated useful signal.