Offload the Overhead, Maintain the Craft

Brett Christie, Ph.D.
VP, Educational Innovations & Inclusivity

On March 26, 2026, we hosted a webinar focused on a pragmatic shift in how educators can think about AI at work: moving from abstract debates about “AI in education” toward the concrete reality of educator workflows.

Joining me were Casandra Silva Sibilin (York College, CUNY), Zach Justus (CSU, Chico). Together, we explored how “human with AI” workflows can be designed so that routine overhead is reduced while human judgment remains the driver of consequential decisions.

The Framing That Unlocked the Conversation: Bots vs. Agentic AI

A key framing for the webinar was the distinction between bots and agentic AI:

• Bots are best for bounded, repeatable tasks with clear triggers and rules (routing, reminders, formatting, checks).
• Agentic AI can support more goal-directed, multi-step work by drafting options, sequencing steps, and proposing next moves, with human checkpoints.

Rather than treating these as theoretical categories, we emphasized them as a practical design question: What work should be automated, what work should be proposed as options, and what work must remain human-led?

A Two-Minute Community Brainstorm: “Offload to AI, Keep Human-Led”

To bring the concept down to the day-to-day level, we engaged participants in a Padlet activity with a simple prompt:

“What would you offload to AI, and what would you keep human-led?”

Many of the 250+ participants contributed responses. The value of this prompt is that it lowers the barrier to participation while still surfacing actionable workflow ideas. Participants did not need to design an entire bot or agentic system. They only needed to name one task they would delegate to AI support, and one task they would intentionally keep human-led to preserve quality, control, and care.

Even in a quick activity, clear patterns emerged.

What Participants Shared: The Main Themes

Theme 1: Drafting and preparation, especially when structure is known

A large share of responses focused on offloading “first-pass” work where the structure already exists, and where the human role is to refine, contextualize, and decide. One post captured the pattern cleanly:

“Offload to agentic AI: Draft weekly announcement from template.”
“Keep human-led: Personalize emphasis.”

A participant also described building structured planning support into a custom Lesson Planner tool, reinforcing that planning is where many educators see near-term gains with low risk.

Interpretation: Educators want AI to accelerate drafting and organization, while keeping the instructor’s voice and pedagogical intent firmly in human hands.

Theme 2: Sensemaking from discussion, without outsourcing instructional presence

Discussion workflows came up repeatedly as both high-volume and high-opportunity. The most consistent “offload” here was synthesis, not judgment. One post made the boundary explicit:

“Offload to agentic AI (Agentic): Summarize weekly discussion themes.”
“Keep human-led: Respond to misconceptions.”

A related comment expands the same idea in operational terms:

“Deal with the massive amount of discussions and signal when the instructor needs to clarify, add more insights, or redirect.”

Interpretation: Participants are not asking AI to “teach the discussion.” They want help summarizing patterns and surfacing signals, so the instructor can respond more strategically.

Theme 3: Assessment and feedback support, with a clear boundary around subjectivity

Assessment and feedback ideas were common, with a consistent constraint: AI can help generate drafts, detect patterns, and align to rubrics, but grading judgments and nuanced feedback remain human-led. One participant captured this succinctly:

“Keep Human: Anything which could be subjective in grading.”

Others referenced rubric checks, objective grading components, and first-pass feedback drafts grounded in criteria.

Interpretation: This is an assessment validity instinct. People want AI to reduce repetition and increase consistency, but not to replace instructor judgment where evaluation is interpretive.

Theme 4: Quality assurance and alignment checks

A significant cluster of responses focused on QA tasks that are essential but time-consuming, especially in online course design and program alignment. Responses referenced checking course alignment maps, mapping competencies to assignments, and validating links and course-document consistency. One comment offered a crisp, practical version of this:

“Offload: specific info validation, URL checking, etc., across course documents.”

Interpretation: Educators see clear value in bot-like automation for “checking and verifying,” and agentic support for proposing structured revisions, while humans remain responsible for final design decisions.

Theme 5: Accessibility and compliance support, with human verification

Accessibility appeared as its own theme, with participants interested in AI support for scanning and detection, and humans retaining responsibility for accuracy and final fixes. A representative post:

“Offload to AI: Scan images for alt text.”

This theme also included “PDF accessibility” checking and correcting errors.

Interpretation: Participants are drawn to AI as a “compliance accelerator,” but they still anticipate human review to confirm appropriateness, accuracy, and compliance to accessibility standards.

Cross-cutting insights

Across categories, the same design pattern showed up again and again:

• Offload to bots: routing, sorting, reminders, checks, validation, formatting
• Offload to agentic AI: drafting options, synthesizing patterns, proposing next steps
• Keep human-led: final decisions, tone and relationship, subjective evaluation, sensitive interactions, escalation pathways

In other words, participants were not seeking “AI takeover.” They were articulating a workable model of human-led, AI-supported practice that reduces overhead while protecting instructional judgment and care.

A Constructive Next Step

If there is one takeaway from the community, it is that educators are ready for a constructive middle ground. They want AI to carry routine workload, they want human judgment to steer, and they want workflows that are transparent enough to defend.

A simple question can initiate real redesign work on a campus team:

What would we offload to AI support, and what will remain human-led by design?