Home Use Cases Work Insights About Contact
Back to Our Work

How a Growth-Stage Technology Company Built Courses That Change Because of Learners, with Instructor Approval

Operations3
Open the walkthrough

Six-Step AI Workflow with Human Oversight

Each step shows the workspace, what the AI completed, and what still needs human review.

Steps 5 and 6 need a person. Step 5 is waiting for approval right now. Click it to open the full walkthrough.

Estimates are directional and based on stated assumptions. All names, organizations, and identifying details have been anonymized in accordance with our confidentiality agreements.

The Transformation

How a Growth-Stage Technology Company Built Courses That Change Because of Learners, with Instructor Approval

Before
An instructor used to start every new learner on the same week-one material.
An instructor used to teach one lesson at one pace to whoever was in the room.
A learner used to save their questions for the weekly live session.
An instructor used to grade every submission by hand, days after it was handed in.
After
every new learner mapped, a path of their own
every exercise at the level the learner can climb from
stuck learners guided, never handed the answer
graded in minutes, feedback that names the step
who is slipping, and the check-ins that carry your name
Executive Summary

Every learner on the platform, from the day they arrive to the day the course changes because of them. The agents map what each new learner already knows, serve every lesson at the level that learner can climb from, guide them when they are stuck, and grade their work in minutes instead of days. A person is handed exactly two things: the learner who is slipping, and the change to the course itself. Six stations, run as a loop: the exercise station 6 re-rates on Friday is the one station 2 serves on Monday, at the new level.

What Was Broken

An instructor used to start every new learner on the same week-one material
An instructor used to teach one lesson at one pace to whoever was in the room
A learner used to save their questions for the weekly live session
The real cost
An instructor used to grade every submission by hand, days after it was handed in.

What We Built

Six stations and 12 subagents. Each subagent carries its own tasks and its own refusal.

A4
Gap Mapper
Check every placement answer against the skill it tests
A1
Path Builder
Draft a starting path from the gaps found, earliest missing skill first
A4
Difficulty Setter
Serve each next exercise off the learner's last ten answers, not their last one
A3
Engagement Watch
Check each session against that learner's own usual length and pace
A1
Tutor
Explain the concept behind the stuck exercise in plain words, off the lesson the learner just read
A3
Stuck Pattern
Match each question to the exercise and the concept it came from
A3
Grader
Grade each submission against the rubric its exercise carries
A1
Feedback Writer
Write the feedback off the specific error, naming the step where the work turned
A4
Progress Tracker
Check every learner's week against their path: cleared, looped back, or gone quiet
A1
Outreach Drafter
Draft the check-in for each flagged learner, off what they were last working on
A4
Cohort Patterns
Check every exercise against the week's attempts: cleared, failed, abandoned
A2
Rating Proposal
Propose the re-rating for each exercise the week's attempts contradict

How It Runs

1

Placement Mapping

Nobody is asked anything here

Agents read every placement answer, trace each miss to the earliest skill it rests on, and build each learner a path that starts where their knowledge actually stops. Nothing waits on a person here. A learner whose answers are too few to map is put on a short probe set, never on a guess.

2

Lesson Pacing

Nobody is asked anything here

Agents serve each next exercise off the learner's own recent record, move the level up or down as the answers come in, and log every session against that learner's own usual rhythm. Nothing waits on a person here. A level is never changed on one bad evening, and a quiet holiday week is marked unscored, not disengaged.

3

Guided Help

Nobody is asked anything here

Agents take each question mid-lesson, walk the learner through a worked example a step at a time, and count every wall that more than one learner hits. Nothing waits on a person here. A learner still stuck after three guided attempts reaches their instructor with the whole transcript, not with a summary.

4

Work Grading

Nobody is asked anything here

Agents grade each submission against its rubric the minute it lands, write feedback that names the step where the work turned, and post both to the learner's record. Nothing waits on a person here. A submission the rubric cannot score reaches an instructor ungraded and marked why, never scored on a guess.

5

Progress Watch

A person answers here

Agents move each learner along their path or loop a failed skill back with fresh exercises, and draft a check-in for every learner gone quiet past their own usual rhythm. This is where a person works. A message to a struggling learner goes out in a person's name, and a learner whose own note explains their silence is never sent a template.

6

Course Tuning

A person answers here

Agents read the week's attempts across the whole cohort, propose a re-rating for every exercise the attempts contradict, and flag the exercise the strong learners fail too. A person accepts or rejects each change. What the course teaches, and at what level, is not something an agent changes quietly.

Where a Person Decides

Step 5, Progress Watch. Send the four drafted check-ins, they carry your name. Write to Amara yourself, or pause her path until her shifts settle. outreach carries your name.
Step 6, Course Tuning. Accept or reject the four re-ratings. Decide whether the subquery exercise comes out tonight or gets rewritten first. the course changes only on your yes.

Operating Model

This changes how work flows through the team.

Role
Responsibility
Progress Watch owner
Send the four drafted check-ins, they carry your name. Write to Amara yourself, or pause her path until her shifts settle. outreach carries your name.
Course Tuning owner
Accept or reject the four re-ratings. Decide whether the subquery exercise comes out tonight or gets rewritten first. the course changes only on your yes.
Placement Mapping, when it goes wrong
Look here only if a learner says their course started in the wrong place.
Lesson Pacing, when it goes wrong
Look here only if a learner says the work got too hard too fast.

What Transfers, What Must Be True

What transfers
A person is in the loop wherever a message goes out in their name or the course itself changes. Two stations refuse to proceed on their own, station 5, Progress Watch, and station 6, Course Tuning, and those are the two a person carries. Everywhere else the agents place, pace, guide and grade, and reach a person only with the learner who is slipping or the exercise that is broken.
Progress Watch stops for a person, and outreach carries your name.
Course Tuning stops for a person, and the course changes only on your yes.
Every subagent says what it will not do. 12 of them do.
What must be true in your environment
The agent can read the systems your records already live in. This one reads 20.
Somebody owns Progress Watch and has time for it.
Somebody owns Course Tuning and has time for it.

Failure Modes

What breaks this pattern:

✗ Guessing a thin placement

A learner answers a handful of questions and the map fills in the rest. They start a month too high or too low, and most learners who start wrong quit before anyone finds out why.

✗ The tutor hands over answers

A tutor built to be helpful gives the answer to graded work. The learner passes the exercise, keeps the gap, and the record now says they know something they do not.

✗ A guessed grade spreads

The grader scores a submission the rubric cannot read, and that wrong grade enters the record. Every pacing and path decision after that builds on a grade nobody checked.

✗ Outreach in nobody's name

An automated message goes to a learner whose own note already said why they stopped. The learner reads it, sees that nobody read the note, and does not come back.

Directional Outcomes

What the agent counts, and the station that counts it.

These counts are the tallies from one monitored run of the agents. They are not monthly or annual totals.

Learners placed
Counted at Placement Mapping
2,380
Started on a skill they already held
Counted at Placement Mapping
0
Placements held for the probe set
Counted at Placement Mapping
74
Exercises served at an adjusted level
Counted at Lesson Pacing
58,400
Level changes a person had to reverse
Counted at Lesson Pacing
0
Learners put on the watch list
Counted at Lesson Pacing
194
Questions answered
Counted at Guided Help
18,700
Answers handed to graded work
Counted at Guided Help
0
Learners routed to an instructor
Counted at Guided Help
246
Submissions graded
Counted at Work Grading
41,200
Grades a learner appealed and won
Counted at Work Grading
0
Submissions sent to an instructor ungraded
Counted at Work Grading
118
Our measurement policy: We do not publish precise ROI without baseline methodology. Every figure above carries its basis.

What Runs Where

Every step names the subagent that does the work, the record it writes, the thing that raises a question for a person, and what it is allowed to touch. This is drawn from the source, not from a diagram somebody kept in sync by hand.

1Placement Mapping
subagentplacement-map
writeslearners/<learner-id>/path.json
raisestoo-few-answers-to-map
may touchlearners/**, the lesson library read-only
2Lesson Pacing
subagentpace-lessons
writeslearners/<learner-id>/sessions/<session-id>.json
raisessignal-too-thin-to-adjust
may touchlearners/**, lessons/** read-only
3Guided Help
subagenttutor-chat
writeslearners/<learner-id>/help/<chat-id>.json
raisesstuck-after-three-attempts
may touchlearners/**, lessons/** read-only
4Work Grading
subagentgrade-and-feedback
writeslearners/<learner-id>/grades/<submission-id>.json
raisesrubric-cannot-score
may touchlearners/**, the submission queue read-only
5Progress Watch
GATE
subagentprogress-watch
writeswatchlist/<learner-id>.json, outreach/drafts/<learner-id>.md
raisesawait-your-send
may touchwatchlist/**, outreach/**, learners/** read-only
6Course Tuning
GATE
subagentcourse-tune
writesratings/proposed/<exercise-id>.json, reports/week-<date>.json
raisesaccept-rating-change
may touchratings/proposed/**, reports/**, the attempt log read-only

Stack

Every system this agent reads or writes.

System
Read at
Stations
the drafted check-ins
Progress Watch
1 of 6
the exercise bank
Lesson Pacing
1 of 6
the exercise ratings
Course Tuning
1 of 6
the instructor grading queue
Work Grading
1 of 6
the instructor queue
Guided Help
1 of 6
the instructor's outbox
Progress Watch
1 of 6
the learner record
Placement Mapping
4 of 6
the learner records
Progress Watch
1 of 6
the lesson library
Placement Mapping
2 of 6
the lesson player
Lesson Pacing
1 of 6
the path builder
Course Tuning
1 of 6
the placement quiz
Placement Mapping
1 of 6
the reporting dashboard
Course Tuning
1 of 6
the rubric library
Work Grading
1 of 6
the session log
Lesson Pacing
1 of 6
the skill map
Placement Mapping
1 of 6
the submission queue
Work Grading
1 of 6
the tutor chat
Guided Help
1 of 6
the watch list
Progress Watch
1 of 6
the week's attempt log
Course Tuning
1 of 6
Next Step

Want to see if this pattern fits your edtech?

No build commitment·Real samples, not a demo·Estimate in writing