Machine-First, Human-Second Workflow
Test every suitable brief with AI, then route failures to people and refine successes.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 7
- Confidence
- 98%
Route suitable work through AI before assigning full human production effort. Give the machine the brief, inspect its response, and iterate when prompting can correct the deficiency. Accept outputs that meet the quality threshold and use a person to verify, polish, and add distinctive expertise. When the machine cannot produce an adequate result, pass the task and any useful preliminary material to a human. Measure how many briefs succeed at the machine stage and gradually update the routing ratio as models, prompts, and team skills improve. In content workflows, apply AI especially to ideation, outlining, and research so writers can devote more attention to composition, voice, judgment, and the improvements that make a final piece exceptional.
Origin
Extracted from Marketing Against The Grain, where the hosts proposed running all startup briefs through AI, retaining successful outputs, and handing failed cases to humans.
Core principles
- 01Use the faster and cheaper machine as the first execution layer.
- 02A failed AI attempt supplies information before human work begins.
- 03Route work according to demonstrated output quality rather than assumptions.
- 04Keep successful machine outputs in the workflow and escalate unsuitable work to people.
- 05Shift the machine-to-human allocation as capability improves.
- 06Human effort creates the most value in judgment, refinement, and difficult edge cases.
How to run it
- 1
Identify Suitable Work
Choose briefs that can safely be attempted by AI and whose outputs can be evaluated before use. Define any privacy, factual, or compliance boundaries before sending material to the system.
Pro tip Begin with reversible internal tasks such as ideation, research questions, and outlines.
Watch out Do not submit sensitive inputs merely because the machine is the default first step.
- 2
Set Acceptance Criteria
Specify what a usable response must contain and which failures require human escalation. Make quality observable enough that the team can route work consistently.
Pro tip Use criteria for accuracy, relevance, differentiation, completeness, and voice.
Watch out Without a quality threshold, teams may accept fluent but inadequate output.
- 3
Run the Machine Attempt
Submit the brief to the AI before committing full human production time. Preserve the response even if it is incomplete because parts may support later work.
Pro tip Use a well-structured creative-brief prompt rather than a one-line request.
Watch out Machine-first does not mean machine-only.
- 4
Iterate Correctable Defects
If the response is close to acceptable, refine the instructions and ask the AI to repair the weak sections. Stop when further prompting costs more than human intervention.
Pro tip Target each revision at a named defect instead of regenerating everything.
Watch out Endless iteration can erase the speed advantage.
- 5
Route by Demonstrated Quality
Keep outputs that pass the criteria and escalate failures to a person. Give the human any usable research, options, or structure created during the machine attempt.
Pro tip Separate complete failures from partial outputs that can still shorten human work.
Watch out Do not force a machine output through solely to improve automation metrics.
- 6
Apply Human Refinement
Have a person verify facts, exercise judgment, strengthen the voice, and add the final differentiating layer. Focus human time on the parts where expertise has the highest leverage.
Pro tip For content, reserve significant attention for the actual writing and final argument.
Watch out A polished draft can still require source verification and editorial accountability.
- 7
Shift the Routing Ratio
Track the proportion of machine attempts that become usable and the defects that repeatedly trigger escalation. Improve prompts and move more work to the machine only when measured results justify it.
Pro tip Review the routing ratio by task type rather than applying one percentage to every workflow.
Watch out Do not assume capability improvements affect all tasks equally.
In the wild
A startup routes every suitable creative brief through AI. Twenty outputs meet its initial standard and proceed to human verification and polish; the other eighty are assigned to people, accompanied by any useful ideas or research produced by the machine. The team records why each failure occurred.
→ The startup saves effort immediately on successful briefs and builds evidence for improving the machine's share over time.
A writer uses AI for ideation, outlining, and preliminary research before beginning a high-value article. Instead of spending most of the schedule assembling a starting point, the writer verifies the material and invests the recovered time in argument, voice, examples, and editing.
→ The workflow accelerates preparation while increasing the human attention available for the final writing.
Common mistakes
Assuming Machine-First Means Publish-First
Routing a brief through AI first does not authorize publishing its answer without human evaluation, verification, and refinement.
Escalating Before Iterating
A near-usable response may only need a clearer prompt or focused revision. Immediate escalation can waste the learning and speed available from the machine stage.
Automating the Highest-Risk Work First
The workflow should begin with safe, reversible tasks whose outputs can be inspected. Sensitive or irreversible work may require a different routing rule.
Is it for you?
Best for
It is best for teams processing repeated research, ideation, outlining, writing, or creative briefs that can be evaluated before publication.
Not ideal for
It is not ideal for sensitive or irreversible tasks where even an initial machine attempt creates unacceptable privacy, safety, or compliance risk.
From the transcript
“If you don't know, start with a machine. Machine's faster and cheaper.”
“Yeah, machine first, humans second.”
“go through the AI first and the human as the kind of sugar on top.”
From the episode
Winners and Losers in the Next Generation of Artificial Intelligence