Why AI Prompts Are Not Enough and Why the Focus Should Be on Workflows Instead

Ever have AI do something impressive once, and then when you tried it again it either failed, gave a completely different answer, or became too complicated to use consistently?

That is a common experience for teachers trying AI for the first time.

A lot of educators assume:

• they are prompting incorrectly
• they need better prompts
• or they just need more practice using AI

But most of the time, that is not actually the problem.

The bigger issue is that people are trying to use AI as a one-off tool instead of building it into a workflow.

That distinction matters.

---

The Prompt Is Not Really the System

AI prompting can be complicated and highly dependent on the model being used.

ChatGPT, Gemini, and Copilot all respond differently. Even different versions of the same model will interpret prompts differently because they were trained differently and predict responses differently.

That is one of the inherent limitations of these systems.

But there is an important pattern here:

AI usually performs best when:

• the task is clear
• the context is specific
• and the expected output is easy to verify

That is why prompts like:

• “Summarize this dataset”
• “Identify the topic of this paragraph”
• “Convert this PDF into a document”
• “What is 4 + 4?”

tend to work consistently.

The task is narrow, the context is clear, and the output can be checked quickly.

But something broader like:

“Make me a sub plan for photosynthesis”

can produce wildly different outputs between models and even between attempts on the same model.

That is because the AI is missing the larger instructional context.

---

The Real Question Is Not “What Prompt Should I Use?”

The better question is:

Where does this fit into my workflow?

That changes everything.

Most teacher workload problems are not single-task problems.

They are workflow problems.

That means the goal should not just be:

• getting one good output

The goal should be:

• reducing repetitive work
• improving consistency
• organizing information
• and supporting decision-making across the larger instructional process

Once you identify the workflow, then you can design the prompts and context around it.

Not the other way around.

---

A Practical Way to Think About AI Tasks

One of the easiest ways I have found to evaluate AI tasks is to ask:

Who could realistically do this task for me?

---

Level 1: Could a Student TA Do This?

These are repetitive and structured tasks that do not require much reasoning.

Examples:

• Converting a PDF into a document
• Matching answers to an answer key
• Organizing information into a spreadsheet
• Reformatting text
• Creating simple practice questions

AI is usually very good at these tasks because:

• they are structured
• they require limited context
• and the output is easy to verify

The prompts also tend to be simpler.

---

Level 2: Could a Teacher Aide or Student Teacher Do This?

These tasks require more reasoning and guidance, but still operate inside a structured system.

Examples:

• Creating vocabulary quizzes
• Identifying Tier 2 and Tier 3 vocabulary
• Drafting differentiated directions
• Grading short responses using a rubric
• Creating first drafts of instructional materials

This is where context and guardrails become much more important.

The AI needs:

• examples
• constraints
• source material
• expected output formats
• and a clearly defined role

The more structure you provide, the more reliable the output becomes.

---

Level 3: Could a Trusted Co-Teacher Help With This?

These are higher-level instructional tasks that require significant context and human oversight.

Examples:

• Analyzing pre-assessment data
• Identifying learning gaps
• Reviewing IEP accommodations
• Building differentiated assignments
• Adjusting instruction based on student performance

AI can support these tasks, but this is where many systems start to fail if they do not have enough context.

The model may need:

• unit goals
• pacing information
• accommodations
• available instructional resources
• assessment data
• classroom constraints
• and the intended outcome of the lesson or unit

Without that context, the outputs often become generic or inconsistent.

This is also where human review becomes critical.

---

Why Workflow Matters More Than Prompting

A good workflow reduces the amount of thinking teachers have to repeat.

That is where AI becomes most valuable.

Not because it is “intelligent.”

But because it can help manage repetitive systems and structured tasks when given the right context.

The prompt itself is usually not the real solution.

The workflow is.

Once the workflow is clear:

• the prompts become easier
• the outputs become more consistent
• and the system becomes more sustainable long term

---

Final Thought

There is a lot AI can help with in education.

But the amount of context, structure, and human oversight needed depends entirely on the task and where it fits into the instructional workflow.

That is why I do not think the focus should be on learning “perfect prompts.”

I think the focus should be on building better workflows first.

Then designing AI support around those workflows in ways that actually save time and improve consistency without removing teacher judgment.