Before You Hire, Audit What AI Can Handle

The expensive mistake is hiring before you understand the work

A job title is not a workflow.

When your business needs more capacity, the reflex is often to open a role, write a job description, and add a person.

That may be the right decision.

Making it before you have mapped the work means you are paying for a bundle of tasks you have not examined.

In the video, I react to a real LinkedIn opening for a web content writer. My reaction is blunt: before committing to a full-time hire for recurring digital work, find out what AI can already handle.

A public job post does not reveal the company’s internal systems, risks, or reasons for hiring. It is not proof that the company made a bad decision.

The useful question is:

Have you tested the work before deciding the work requires another person?

Stop asking whether AI can replace the role

“Can AI replace a web content writer?” is too blunt to be useful.

A role is a bundle of tasks:

• Audience and customer research
• Source gathering
• Topic selection
• Expert interviews
• Outlining
• Drafting
• Fact-checking
• Approval
• CMS entry
• Publishing
• Performance review
• Updating old content

Some tasks may be repeatable, digital, source-based, and easy to check. Others depend on judgment, relationships, context, accountability, or sensitive approval.

The International Labour Organization’s 2025 task-level analysis reinforces that distinction. It assessed nearly 30,000 occupational tasks and concluded that job transformation is more likely than wholesale replacement because most occupations still contain tasks requiring human input. (ilo.org)

The better business question is:

Which tasks can AI perform reliably, which tasks must a person own, and how should the two work together?

Run a pre-hire AI work audit

The video gives the decision rule. The audit below turns it into a usable operating test.

1. Define the outcome you need

Do not start with the job title.

State the business result the work must produce. For web content, the outcome might be:

Publish accurate, useful pages that address real customer questions and create qualified inquiries.

That is specific enough to evaluate.

“Hire a content writer” is not.

2. Break the role into observable tasks

List the work in enough detail that someone can see what enters the process, what happens, what comes out, and who decides whether it is acceptable.

Do not collapse the whole role into “content.”

“Interview an internal expert, turn the transcript into an evidence-backed draft, and place it in the CMS for approval” is a workflow.

“Write content” is not.

3. Identify the strongest AI candidates

A task is worth testing when it:

• Repeats
• Happens inside digital systems
• Can be grounded in approved source material
• Produces an output that can be checked
• Has a reversible failure

These are candidates for a pilot, not automatic permission to automate them.

4. Mark the work a person must own

Keep a human owner where the work requires:

• Business judgment
• Original expertise
• Trust and relationships
• Resolution of conflicting evidence
• Sensitive claims or approvals
• Accountability for the result
• Decisions with a high cost of error

AI can assist with parts of this work.

It should not inherit accountability by default.

5. Build the smallest useful workflow

A real AI agent is more than a chat window with a grand title.

OpenAI describes an agent as a system that can independently accomplish tasks using a model, tools, instructions, and guardrails. OpenAI also recommends validating whether the use case actually needs an agent before building one. (openai.com)

Start narrow.

A valid first workflow could be:

Use an approved source packet and voice guide to create a CMS draft, flag unsupported claims, and stop for human approval before publication.

Do not start with:

Run our entire content operation.

6. Test it on real work

Use representative inputs from your business. Then measure:

• Percentage accepted without a material rewrite
• Human review and editing time
• Factual, brand, or compliance errors
• Cycle time
• Escalation frequency
• Cost per accepted output

Set the acceptance threshold before the pilot.

Otherwise enthusiasm can move the goalposts after a weak result.

7. Make one of three decisions

The test should lead to one of three outcomes:

1. AI workflow: The work is defined, repeatable, reliable, and reviewable.
2. Person: The work still depends heavily on judgment, relationships, accountability, or context.
3. Hybrid: AI handles repeatable production while a person owns direction, exceptions, approval, and the result.

Do not force the answer.

Let the evidence decide.

What an AI content system actually needs

Around-the-clock availability is not understanding.

AI can operate outside office hours. It does not automatically know your audience, business truth, positioning, legal boundaries, or standard of quality.

A useful content system needs:

• An approved source library
• Evidence of customer questions and problems
• Clear positioning and voice rules
• Rules for claims, citations, and prohibited language
• Access to the right tools at the right permission level
• Defined outputs and escalation conditions
• Human review before sensitive or public actions
• A measurement and update loop

OpenAI’s agent guidance treats instructions, tools, guardrails, and evaluation as foundational. NIST’s Generative AI Profile likewise frames trustworthiness as something organizations should build into the design, development, use, and evaluation of AI systems. (openai.com)

That is the difference between using AI to generate words and building a system that can be trusted with business work.

A practical split for web content

Research

AI can help with: Organize approved sources, summarize material, and surface repeated questions.

A person owns: Decide which problems matter and which sources are credible.

Planning

AI can help with: Propose topic maps, outlines, and update opportunities.

A person owns: Set priorities, positioning, and business relevance.

Creation

AI can help with: Draft from approved evidence and specified instructions.

A person owns: Add original expertise, interview people, verify facts, and approve the final voice.

Publishing

AI can help with: Prepare CMS fields, metadata, links, and a review-ready draft.

A person owns: Approve claims, sensitive language, and publication.

Improvement

AI can help with: Group performance data and flag stale or incomplete pages.

A person owns: Decide what changes, what gets removed, and why.

The split may change by workflow.

The ownership principle should not.

When you should still hire

A hire remains the right answer when:

• The work depends on trust or ongoing relationships
• The value comes from original judgment
• The source material is sensitive, inconsistent, or difficult to verify
• Errors create meaningful legal, financial, brand, or customer risk
• The AI pilot repeatedly misses the required quality or efficiency threshold
• The remaining work needs accountable capacity that cannot be reduced to a reliable process

Do not use AI as an excuse to avoid a needed hire.

Use the audit to make the hire more precise.

The decision before adding payroll

Before opening the role, answer these questions:

1. What outcome are we buying?
2. What tasks create that outcome?
3. Which tasks can AI perform reliably?
4. What data, instructions, and tools does AI need?
5. Where is human judgment or approval mandatory?
6. What did the pilot prove?
7. What work remains after the workflow is built?

Then hire for the remaining gap.

Build AI around the repeatable work.

Keep accountability with a person.

Frequently asked questions

Should you automate before hiring?

You should map and test the work before hiring for recurring digital tasks.

That does not mean every task should be automated. The audit gives you evidence for deciding whether the right answer is a person, an AI workflow, or both.

What is an AI agent?

An AI agent is a system that uses a model, tools, and instructions to complete tasks with some independence while operating within defined guardrails.

It is not the same as asking a chatbot to produce one response. (openai.com)

Can an AI agent replace a content writer?

An AI system may handle parts of research, planning, drafting, CMS preparation, and performance review.

It does not automatically possess your company’s judgment, source quality, positioning, relationships, or accountability. The right answer depends on the tasks and the result of a real pilot.

What is a pre-hire AI audit?

A pre-hire AI audit breaks a proposed role into tasks, identifies which tasks are plausible candidates for AI, tests them on real work, defines human approval, and measures what remains before the business commits to hiring.

What should you measure in the pilot?

Measure accepted output, editing time, factual and brand errors, cycle time, escalations, and cost per accepted result.

Choose the threshold before the pilot begins.

Map the work before you hire

Fargo Factor helps founders and business owners map recurring work, decide where AI belongs, define human approval, and design the operating system around the work.

The goal is not fewer people at any cost.

The goal is to stop adding payroll around work that was never designed.

Sources

• Jeff Fargo, Fargo Talks: Before you hire someone, what can AI do?
• International Labour Organization: Generative AI and Jobs: A Refined Global Index of Occupational Exposure
• OpenAI: A Practical Guide to Building Agents
• National Institute of Standards and Technology: Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile

Jeff Fargo is the founder of Fargo Factor and host of Fargo Talks. Fargo Factor advises founders and business owners on practical AI integration, decision systems, operating systems, marketing systems, and AI search visibility.