The AI-Era Three-Part Talent Filter
Hire for deep expertise, AI orchestration, or high-trust human work
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 97%
Steven Bartlett says his own hiring behavior has changed as internal AI agents can perform work he previously assigned to people. He groups the candidates he currently values into three categories: specialists with deep expertise, AI-proficient operators who can redesign workflows and manage agents, and people with strong human-to-human skills for relationship-dependent work such as some sales. The reusable filter starts with the role's required outcome, tests which parts tools can already perform, and then asks which of the three scarce contributions remains necessary. A realistic work sample checks the judgment rather than relying on labels. This is Bartlett's current operating hypothesis, not established labor-market science, and the transcript also contains counterexamples suggesting demand for programmers may grow as producing technology becomes cheaper.
Origin
Steven Bartlett describes using this filter in his companies; extracted from The Diary of a CEO.
Core principles
- 01A role should add a capability that tools do not already supply
- 02Deep expertise remains valuable when judgment is hard to automate
- 03AI-proficient operators can multiply output by managing agents
- 04Some high-trust relationships still benefit from human presence
- 05A hiring filter is a hypothesis to test, not proof that other workers lack value
How to run it
- 1
Define the owned outcome
Describe the result the person must produce and the decisions they must make. Remove inherited tasks that no longer contribute to that outcome.
Pro tip Use a measurable deliverable from the first 90 days.
Watch out Do not begin with a legacy job title.
- 2
Test the tool baseline
Run the current workflow with available AI and automation before opening the role. Record where the tools fail, require supervision, or create unacceptable risk.
Pro tip Have the team use real inputs rather than a polished demo.
Watch out A successful demo does not prove reliable production performance.
- 3
Score specialist depth
Ask whether the remaining work requires hard-won domain judgment, original initiative, or accountability that the tools and current team lack. Verify it through a domain-specific work sample.
Pro tip Look for decisions and error detection, not years of experience alone.
- 4
Score AI orchestration
Assess whether the candidate can design workflows, supervise agents, evaluate outputs, and recover from failures. Test them on the actual systems the team uses.
Pro tip Include a deliberately flawed agent output in the exercise.
Watch out Tool fluency without domain judgment can scale errors.
- 5
Score human trust
Determine whether success depends on rapport, negotiation, reassurance, physical presence, or personal accountability. Test those interactions rather than assuming every customer wants a human.
Pro tip Ask customers which moments require a person and why.
Watch out The transcript offers sales as an example, not a permanent boundary around human work.
- 6
Choose and review
Hire when the candidate clearly fills at least one important gap and can work with the rest of the system. Recheck the role after tools and workflows change.
Pro tip Record the capability gap that justified the hire.
Watch out Do not use the filter to erase junior development paths without testing alternatives.
In the wild
Bartlett says his investment fund stopped interviewing for additional analyst roles after concluding that one existing analyst could operate with AI. He reports that she had set up three agents that performed work he associates with three people.
→ The company chose to augment an existing specialist rather than add the previously considered roles.
A company automates research and proposal drafting but finds enterprise buyers still need a named person to negotiate trade-offs and accept responsibility. It hires a relationship-led salesperson who uses the tools rather than duplicating their output.
→ Automation handles preparation while the hire owns trust, negotiation, and accountability.
Common mistakes
Using categories as permanent labels
The filter assesses the capability required by a role now; people can develop expertise and AI proficiency.
Trusting an AI demo
Production work needs testing for reliability, supervision cost, and consequences of error.
Eliminating junior pathways blindly
The episode describes current hiring pressure but does not solve how organizations will develop future experts.
Is it for you?
Best for
It is best for teams redesigning roles while AI agents absorb parts of research, analysis, coding, or administration.
Not ideal for
It is not ideal as a blanket reason to reject junior candidates or assume that relationship work and expert judgment cannot change.
From the transcript
“I'd say it's three groups.”
“People that have very, very deep expertise on a particular thing”
“People that are AI proficient”
From the episode
Financial Crash Expert: In 3 months We’ll Enter A Famine! If Iran Doesn’t Surrender It's The End!