Home Use Cases Work Insights About Contact
Back to Our Work

How a Growth-Stage Technology Company Built Learning Paths Where Difficulty Moves with the Learner

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 Learning Paths Where Difficulty Moves with the Learner

Before
An instructor used to give every new learner the same forty-question test and mark it by hand.
An instructor used to hand every learner the same module list, whatever they already knew.
A teaching assistant used to answer stuck learners by ticket, days after the session ended.
An instructor used to grade submissions once a week and send feedback days after the miss.
After
every new learner measured, the contradictions held
a path per learner, sized to their week, no gap papered over
difficulty moving with each learner, stuck steps coached live
graded the same day, the edge cases held for a person
six learners on one desk, drafts written, your call waiting
Executive Summary

Every learner, from the day they enrol to the day they finish, on a path built from what they actually know. The agents place each learner, build their path, move the difficulty with them, answer their stuck questions, grade their work the same day, and put on an instructor's desk only the learners whose next step carries a person's name: an outreach message, a borderline final, a change to the course itself. Six stations, run as a loop: the step that stalled eleven learners this week is the module station 2 builds paths around next week.

What Was Broken

An instructor used to give every new learner the same forty-question test and mark it by hand
An instructor used to hand every learner the same module list, whatever they already knew
A teaching assistant used to answer stuck learners by ticket, days after the session ended
The real cost
An instructor used to grade submissions once a week and send feedback days after the miss.

What We Built

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

A4
Skill Check
Give each new learner the placement set for the course they joined
A3
Gap Finder
Locate the skills each learner already holds, from their answers alone
A3
Module Match
Match each missing skill to the module that teaches it, not the module nearest it
A4
Load Setter
Set each learner's starting difficulty from their own placement, not the cohort
A4
Difficulty Dial
Track each learner's answers, pace and retries inside the session as they work
A1
Step Tutor
Explain the step a learner is stuck on, from the worked example behind it
A3
Rubric Grader
Grade each submission against the rubric its module carries
A1
Feedback Writer
Draft feedback for each graded submission, naming the step that went wrong
A4
Risk Watch
Score each learner's week on pace, misses, session gaps and tutor asks
A1
Outreach Drafter
Draft the message to each flagged learner from their own record, in the instructor's voice
A4
Stall Finder
Check every module against where learners stalled, quit, or asked the tutor the same thing
A2
Change Proposal
Propose the content or sequence change that would have removed each repeated stall

How It Runs

1

Learner Placement

Nobody is asked anything here

Agents give every new learner an adaptive placement, read the answers into held, shaky and missing skills, and hold any placement whose answers contradict each other. Nothing waits on a person here. A held placement is never averaged into a starting level; it reaches the instructor's desk with its answers attached.

2

Path Building

Nobody is asked anything here

Agents build each learner's path from the modules that teach their missing skills, size it to the hours they gave, and flag any gap no module covers instead of assigning the nearest one. Nothing waits on a person here. A gap with no content behind it goes to course tuning, not into a path that pretends to cover it.

3

Session Coaching

Nobody is asked anything here

Agents move each learner's difficulty on their answers inside the session, coach the step a learner is stuck on, and refuse to hand over the answer to anything graded. Nothing waits on a person here. A learner who stalls gets the step explained in the session, not a ticket answered next week.

4

Submission Grading

Nobody is asked anything here

Agents grade every submission against its rubric the day it lands, write feedback that names the step that went wrong, and hold every grade that sits on an edge a person should call. Nothing waits on a person here. A held grade reaches the instructor's desk at station 5; everything clear of the edges releases with its feedback the same day.

5

Instructor Briefing

A person answers here

Agents score every learner's week against their own baseline, put the sliding ones on one desk with the held placements and held grades beside them, and draft the outreach in the instructor's voice. This is where a person works. Everything else exists so that six learners, not two hundred and thirty-eight, reach this desk.

6

Course Tuning

A person answers here

Agents trace every stall, quit and repeated tutor question back to the step that caused it, propose the content change that would have removed it, and propose nothing where one learner stalled once. A person accepts or rejects each change. What the course teaches, and in what order, is not something an agent changes quietly.

Where a Person Decides

Step 5, Instructor Briefing. Send the three messages, or edit them first. Call Priya's final, the matching answers, and Mei-Ling's placement. your name, your learners, your call.
Step 6, Course Tuning. Accept or reject the two proposed changes. Say whether module 2 gets rewritten this term or next. the course changes on your say.

Operating Model

This changes how work flows through the team.

Role
Responsibility
Instructor Briefing owner
Send the three messages, or edit them first. Call Priya's final, the matching answers, and Mei-Ling's placement. your name, your learners, your call.
Course Tuning owner
Accept or reject the two proposed changes. Say whether module 2 gets rewritten this term or next. the course changes on your say.
Learner Placement, when it goes wrong
Look here only if a learner says their placement felt wrong.
Path Building, when it goes wrong
Look here only if a learner asks why their path skips a module.
Session Coaching, when it goes wrong
Look here only if a learner says the level keeps fighting them.
Submission Grading, when it goes wrong
Look here only if a learner disputes a released grade.

What Transfers, What Must Be True

What transfers
a person is in the loop wherever a message goes out in their name, a grade releases on an edge, or the course itself changes. Two stations refuse to proceed without a person, Instructor Briefing and Course Tuning; everywhere else the agents place, build, coach and grade, and reach the instructor only when the call is theirs to make.
Instructor Briefing stops for a person, and your name, your learners, your call.
Course Tuning stops for a person, and the course changes on your say.
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 Instructor Briefing and has time for it.
Somebody owns Course Tuning and has time for it.

Failure Modes

What breaks this pattern:

✗ Averaged answers place learners wrong

When a placement contradicts itself and the system averages it anyway, the learner starts a course at a level nobody actually measured. They spend weeks bored or drowning before anyone notices the placement was fiction.

✗ Paths stretch past the deadline

When a path cannot finish inside the learner's window and the system stretches it instead of saying so, the learner finds out at the deadline, not at enrollment. By then the time to renegotiate is gone.

✗ The tutor sells the answer

When the tutor hands over the answer to a graded question, the grade stops meaning the learner can do the work. Every certificate downstream inherits that doubt.

✗ Edge grades get rounded silently

When a score one point off a pass line gets rounded instead of held, a machine changes someone's transcript and no person ever sees the evidence. Matching answers slip through the same gap, so copied work passes ungraded.

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.

new learners placed since Friday
Counted at Learner Placement
12
contradictory placements averaged into a level
Counted at Learner Placement
0
placements held for the instructor's desk
Counted at Learner Placement
2
paths built from this morning's placements
Counted at Path Building
11
gaps papered over with the nearest module
Counted at Path Building
0
gap without content sent to tuning
Counted at Path Building
1
questions coached from the worked examples
Counted at Session Coaching
61
graded answers handed to a learner
Counted at Session Coaching
0
stall pattern sent to course tuning
Counted at Session Coaching
1
submissions graded and released same day
Counted at Submission Grading
136
edge grades rounded across a line
Counted at Submission Grading
0
grades held for a person's call
Counted at Submission Grading
3
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.

1Learner Placement
subagentplacement
writeslearners/<id>/placement.json
raiseshold-contradictory-placement
may touchlearners/*/placement.json, the question bank and skill list read-only
2Path Building
subagentpath-builder
writeslearners/<id>/path.json
raisesgap-without-content
may touchlearners/*/path.json, content/modules/** read-only
3Session Coaching
subagentsession-floor
writeslearners/<id>/sessions/<date>.json
raisesgraded-answer-requested
may touchlearners/*/sessions/**, content/** read-only
4Submission Grading
subagentgrader
writeslearners/<id>/grades/<submission-id>.json
raiseshold-borderline-grade
may touchlearners/*/grades/**, rubrics/** read-only
5Instructor Briefing
GATE
subagentintervention-desk
writesdesk/<week>/<learner-id>.md
raisesawait-instructor-call
may touchdesk/**, everything else read-only
6Course Tuning
GATE
subagentcourse-tuning
writeschanges/proposed/<change-id>.json, reports/course-<date>.json
raisesaccept-course-change
may touchchanges/proposed/**, reports/**, the history read-only

Stack

Every system this agent reads or writes.

System
Read at
Stations
the change log
Course Tuning
1 of 6
the course completion rules
Path Building
1 of 6
the course skill list
Learner Placement
1 of 6
the enrolment list
Learner Placement
1 of 6
the gap flags from path building
Course Tuning
1 of 6
the gradebook
Submission Grading
2 of 6
the held items from stations 1 and 4
Instructor Briefing
1 of 6
the learner record
Learner Placement
4 of 6
the learner records
Instructor Briefing
1 of 6
the message drafts
Instructor Briefing
1 of 6
the module catalog
Path Building
2 of 6
the module content
Session Coaching
1 of 6
the module rubrics
Submission Grading
1 of 6
the placement question bank
Learner Placement
1 of 6
the prerequisite chart
Path Building
1 of 6
the session player
Session Coaching
1 of 6
the submission queue
Submission Grading
1 of 6
the tutor question log
Course Tuning
1 of 6
the week's session records
Course Tuning
1 of 6
the worked examples
Session Coaching
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