How I Manage AI Like a Project Team: Applying the 4D AI Fluency Framework
AI assistants have different capabilities. Combined correctly, they become valuable members of a project team — but only with the same discipline you would apply to human ones.
Anthropic's 4D framework introduces four core concepts for working effectively with AI in projects: Description, Delegation, Discernment, Diligence. The idea is simple but powerful — AI assistants have different capabilities, and when combined correctly they become valuable members of a project team.
Here is how those principles look when applied to real delivery rather than to demos.
The four dimensions
| Dimension | What it asks of you |
|---|---|
| Description | State what you want, why, and against which acceptance criteria |
| Delegation | Decide which parts belong to the AI, which to you, and which are shared |
| Discernment | Evaluate the output critically instead of accepting it |
| Diligence | Take responsibility for what ships, including verification and ethics |
Description and delegation in practice: a five-step workflow
The workflow below assigns an explicit owner to every step. That single detail — naming the owner — is what turns AI use into AI collaboration.
| Step | Owner | What happens |
|---|---|---|
| 1 · Requirement clarification | Human | Structured product requirements with acceptance criteria and task priorities |
| 2 · Q&A session | Together | The AI is given thinking space and encouraged to ask questions, surfacing missing requirements and clarifying scope |
| 3 · Implementation | AI | The AI generates the code and delivers the implementation files |
| 4 · Testing | Human | Command-line testing for outputs, API/Swagger testing for requests and responses, workflow testing for end-to-end behaviour |
| 5 · Debugging | Together | Screenshots and error logs are shared so the AI can refine the code |
Step 2 is the one most people skip. Giving the model room to ask questions before it writes anything consistently surfaces requirements that were never written down. It costs a few minutes and saves a rewrite.
Steps 4 and 5 are where Discernment and Diligence actually live. Systematic, layered testing is not bureaucracy — it is the mechanism that makes delegation safe.
What human team management already taught us
This working discipline is not new. Anyone who has managed a product or project team already collaborates this way:
- Understanding each member's capabilities
- Observing strengths and weaknesses
- Assigning tasks based on strengths
- Creating opportunities to improve weaker areas
The same leadership mindset transfers directly to working with AI systems. The model has a capability profile; your job is to know it, route work accordingly, and verify what comes back.
Building repeatable success
The goal is not a good session. It is structured collaboration patterns that let successes be repeated across future projects. Those structures evolve continuously through best practices, lessons learned and iterative improvement — which is exactly the organizational memory problem at individual scale.
Working with AI is much like communication practice: the more time you spend collaborating, the more efficient and precise your prompts and conversations become.
Responsible collaboration
For ethical and safe usage, transparency is essential. Define clearly who you are, what you want to achieve, and why you want to achieve it. At the same time, sensitive information must always be protected.
These journeys are still experimental and require patience, creativity and responsibility.
AI systems are powerful — but they are not magical. Humans remain responsible for deployment and decision-making.
When working with AI, we are not only improving machine capabilities. We are also developing our own communication, leadership and systems-thinking skills.
If you are designing this at organizational rather than personal scale, the Agentic RACI framework extends the same ownership logic across teams and agents.
A version of this article first appeared on LinkedIn.
Frequently Asked Questions
What is the 4D AI fluency framework?
Four core concepts for working effectively with AI: Description (stating what you want, why, and against which acceptance criteria), Delegation (deciding which parts belong to the AI, to you, or to both), Discernment (evaluating output critically rather than accepting it) and Diligence (taking responsibility for what ships, including verification and ethics).
How do you structure a human-AI delivery workflow?
Five steps, each with a named owner: requirement clarification (human), a Q&A session where the AI is encouraged to ask questions (together), implementation (AI), layered testing across command line, API and end-to-end workflow (human), and debugging with shared screenshots and error logs (together). Naming the owner at each step is what turns AI use into AI collaboration.
Why give the AI a question-asking step before implementation?
Because it consistently surfaces requirements that were never written down. Giving the model room to ask questions before it produces anything clarifies scope and identifies missing acceptance criteria. It costs a few minutes and typically saves a full rewrite.
Does managing AI resemble managing a human team?
Closely. Both involve understanding each member's capabilities, observing strengths and weaknesses, assigning tasks based on strengths, and creating opportunities to improve weaker areas. The same leadership mindset transfers: know the capability profile, route work accordingly, and verify what comes back.
What does responsible AI collaboration require in practice?
Transparency about who you are, what you want to achieve and why, combined with strict protection of sensitive information. It also requires accepting that AI systems are powerful but not magical — humans remain responsible for deployment and decision-making, which is why systematic verification is part of the workflow rather than an optional extra.
If the answer is not above, ask. We add the questions we receive to this page’s Frequently Asked Questions section.
Ask on LinkedIn Ask by emailIndigonix designs governed enterprise intelligence architectures, agentic systems and AI governance programmes. Talk to us about an assessment, workshop or design engagement.
Contact usSee servicesTrainings & workshops← Indigonix System Intelligence