Starting assumption
AI safety lessons for young adults
My original plan was to complement 
, my career learning platform, with bite-sized AI safety lessons that would give young adults a route from Elm into the AI safety ecosystem.
BlueDot Impact × Fulcrum Challenge
Pathwais
Pathways into AI safety for experienced generalists
Why I pivoted after day one
New direction
Pathways for experienced generalists
Help AI safety teams recognise relevant experience and bring it into senior roles.
I saw easing the bottleneck in generalist hiring as the most meaningful contribution I could make.
Demand for experienced generalists
Training pathways
2,000–2,500
Estimated research fellows in 2026
~300 projected non-research fellows in 2026
5–10 estimated fieldbuilding hires per year from Pathfinder and Astra
In some non-research searches with hundreds of applications, the authors reported only 0–5 applicants meeting core requirements.
Capacity to grow
$71MRaised, with 15 staff at BlueDot
BlueDot estimated it could grow 3–5× in a year if it found the right hires.
Friction in hiring
670Applications for one role
In a separate case, one candidate went through eight hiring stages over eight months.
Mapping a candidate journey into AI safety
Build context
Learn the core AI risks and how organisations address them.
Identify and match in-demand skills
Match your existing skills to roles in AI safety.
Test your fit
Try a small project that resembles the work.
-
Apply
Target relevant jobs and fellowships.
Network
Talk to advisers and people working in the field.
Practise deliberately
Build a portfolio of projects that show what you can do.
Work in public
Share your work and use early feedback to improve it.
Candidate journey resources
Build context
Identify and match in-demand skills
Test your fit
Apply
Network
Practise deliberately
Work in public
Candidate journey
Build context
Identify and match in-demand skills
Test your fit
-
Apply
Network
Practise deliberately
Work in public
What happens after you apply?
MATS · Operations Generalist
- Application
- Screening call
- Paid work sample
- Structured interview
- Longer interview
- References
- Offer
Coefficient Giving
- Application
- Initial screen
- 1–2 paid tests
- Team interviews
- References
- Offer
Apollo Research
- Application review
- 30-minute screen
- 2–2.5-hour task
- Role interviews
- Leadership interview
- Offer
METR · Technical example
- Code screen
- Take-home tests
- Technical assessment
- Behavioural interview
- ~2-day work trial
One application can lead to several rounds of screening, work samples and interviews.
The sequence and time commitment vary by role.
Hiring time by stage
Hiring friction from both sides
Finding a role
Hiring operations
Recruiting through familiar networks
Personal connections were reported as a main route into some roles.
Kairos advertises a $10,000 referral bonus.
Understanding role requirements
Job readiness
Defining the role and criteria
Bundled responsibilities and unfamiliar job language made expectations harder to read.
Showing relevant skills
Job readiness
Interpreting experience from outside AI safety
Some hiring teams reported difficulty assessing external credentials and used familiarity with the field as a signal.
Completing assessments
Job readiness · Hiring operations
Designing role-specific work tests
Teams need to decide what a task should measure and how to score it. The research identified gaps in assessment design and evaluator expertise.
Learning from the decision
Feedback
Providing useful feedback
Interviewees reported limited feedback after hiring decisions.
The criteria draw on the broader Pathwais evidence register, which logged 55 sources, including 4 study sources and 7 individual accounts.
Pathwais · Navigating into and through AI safety careers
Proposal
Pathwais is a proposal to reduce friction in AI safety hiring by connecting candidates and hiring teams with the support they need at each stage. It could take the form of a toolkit, a targeted intervention, a consultancy, an independent body, or infrastructure that connects existing services. Pathwais would build on what already exists and work with organisations in the field to identify and address shared hiring challenges. The aim is to help organisations bring in the people they need sooner, so they can move faster on urgent work within the budgets they have.
Job readiness
Pathwais solution hypothesis
If candidates can log their progress, revisit key learning, and find up-to-date guidance on courses, advisers, and applications, they’ll be more likely to retain what they learn and follow through on their next step.
Hiring operations
Pathwais solution hypotheses
- If teams design work trials around the skills a role actually requires, reviewers will be better able to recognise relevant experience from other fields.
- If teams draw on shared rubrics and assessment examples to tailor their hiring process to each role, they’ll spend less time designing work trials.
Feedback
Pathwais solution hypothesis
If teams give candidates specific feedback after assessments and review recurring themes across hiring rounds, candidates will have clearer next steps and teams will see where to improve the process.
Pathwais toolkit · Feedback
This feedback tool is one example of what the Pathwais toolkit could include. It shows how reviewers could record what a work sample revealed about transferable skills, share feedback with candidates, and improve the next hiring round.
Prototype with fictional data. Open the mobile wireframe
Targeted research to refine Pathwais
200+
Professionals we will speak with and survey
Across the AI safety talent ecosystem
- Experienced generalists
- Recruiting specialists
- In-house recruiters and hiring managers at AI safety organisations
- AI safety talent programme, fellowship and resource leads
- AI safety mentors and career advisers, such as Successif
Kairos / Talent Commons, Leaf, Lateral, Coefficient Giving, Institute for Progress, MATS, Successif and High Impact Professionals.
What I’ll test next
Refine assumptions through targeted feedback
Guiding research questions
- At which stage is candidate attrition highest?
- Which stage costs the hiring team the most to run?
- Which stages require the most active time from candidates and hiring teams?
- Where could we save time while still giving reviewers the evidence they need?
Test the assumptions behind each intervention
Use survey responses and targeted interviews to see which assumptions hold and refine what Pathwais should test first.
Participate in a work trial for a generalist role
Experience the process as a candidate and, where possible, ask reviewers what evidence they needed. Use what I learn to decide which part of the process Pathwais could help improve.