Conjure Assign · The take-home builder

Describe the work.
Get it back graded.

Conjure Assign is the take-home half of Conjure. Write one sentence - a problem set, a case memo, a multi-stage simulation, a client who emails back, a tool students build by describing it - and it's built from your own material into interactive coursework, personalized to every student, and graded against your rubric the moment they hand it in. You read the reasoning. The busywork is already done.

See it in a demo How it works
Conjure · FIN 301 ● Building live
Describe any activity - a spreadsheet, an email thread, an AI agent, a page students build - and Conjure builds it, personalized and auto-graded.
See it in action

Watch Conjure work.

A full walkthrough of Conjure in action — the same builder behind live activities in Lens and take-home coursework here in Conjure Assign. Press play for the tour.

Conjure · Lens — full walkthrough
How it works

One sentence in.
A graded stack out.

No question bank to wire up, no rubric to hand-build. You describe the assignment the way you'd describe it to a TA — Conjure does the rest, and keeps you in the loop at every step.

01
Describe it
Say what you want in plain language, or point Conjure at a reading, dataset, or slide. "A short case memo on this quarter's filing" is enough to start.
02
It builds from your material
Conjure drafts the whole assignment - the questions, the people students will talk to, the tools they will build, the reasoning it expects - grounded in your source, not generic filler. You review and tweak before anything ships.
03
Every student gets their own
Same concept, different context. Each copy is re-skinned to a student's major and interests, so the work is theirs to reason through — and far harder to hand to a chatbot.
04
It grades on submit
The moment a student turns it in, Conjure grades against your rubric, drafts feedback, and flags the answers worth your eyes. You review, adjust, and release.
What it builds with

Not a quiz engine.
A whole vocabulary.

Describe the work and Conjure picks the parts it actually calls for. Auto-checked questions where a rule can mark them; people, agents, tools and files where the learning is in doing the job.

AI conversations
Someone who pushes back
A client, a controller, a patient, a sceptical director - a named character with a position, on the channel the scene would really use: in person, an email thread with a subject line and everyone Cc’d, or a team chat. Several people in one room as a cast. The student’s side is what gets graded.
AI CONVERSATION
FromDana Whitfield · Head of Procurement
SubjectRe: Q3 price change
Twelve percent with two weeks’ notice is not something I can take to my board.
Reply — 2 of 3
AI agents & the Coding Agent
An AI the student directs
An assistant with rules you set - a data assistant that explains any formula but never runs the analysis; a research assistant with a scope; a Coding Agent that builds what they specify, its page and its code beside the thread. Marked on how they used it: what they asked, what they checked, what they refused to accept.
CODING AGENT
Builds what the student specifies · never specifies it for them
build me a fee calculator with a volume tier
Which tier boundary - GMV or units? I’ll write it either way.
fees.html · fees.js
AI page build
A tool they specify into existence
The student describes a calculator, a model, a simple simulator - and it’s built live in front of them. They test it, ask for changes, and hand it over in a conversation with one click. Or the build makes them a file: a spreadsheet with live formulas, a document, a dataset.
AI PAGE BUILD
Prompt 2 of 3 · built live
Break-even calculator
Price$48
Unit cost$19
Break-even 1,242 units
Spreadsheets & files
Work that travels with its formulas
A spreadsheet the student fills in keeps its formulas all the way to grading - the formula is checked, not just the number. Students attach PDFs, Word, Excel and CSV files to a conversation, and the character or agent reads them.
SPREADSHEET
D4fx=D3+B4-C4
MonthCash inBurnLeft
Feb14,50030,00026,500
Mar17,00030,000=D3+B4-C4
Adaptive AI parts
The next question follows from the last
A problem that adapts to how the previous one went - to the student’s level, their major, their earlier answers. It keeps its own state, marks itself, and you set how many tries it allows.
ADAPTIVE AI PART
They got the last one · so this one moves up
Now the receivable is factored at 3%. Does your answer change, and what does the 3% buy?
Attempt 1 of 3 · marks itself
Multi-stage simulations
A career, with consequences
A role at a company across several stages: stage gates where a score bar decides, or a branching ladder where everyone continues and the next stage is set at the level they earned - promoted, or a second chance at the same competencies. Decisions that reshape what comes next.
STAGE GATE
VP of Finance
You’ve been promoted. The board wants the forecast you defended.
78% since stage 2 · bar was 70%
Games, maps & markets
A board to play - and a rival on it
Ask for a game and Conjure builds the thing itself: a market that clears each round, a city map they site depots on, a negotiation that scores, a model they push until it breaks - drawn into the page and playable, not described. Drop an AI player into it and it takes its own turns against them or alongside them, showing one line of reasoning after each move. Nobody chats with it; it plays, and every move is recorded with the rest of their work.
AI RIVAL OR TEAMMATE
VoltCo plays against them, each round
Round 3 · “I took the north corridor - your depot cannot reach it.”
Data & code
Real queries, real code
SQL queries run against tables built for the case, with a model query and an expected result; database diagrams; code views students annotate line by line; predict-the-curve sketches checked against the real one.
SQL QUERY
select region, sum(fees)
from orders group by region
regionsum
Midwest41,200
South38,940
Checked against the model query
Discussion & deliberation
Everyone commits before anyone speaks
Across the whole class, one post everybody answers on their own - and only once they have posted do they see their classmates’ answers and reply to them, with a minimum length, a number of replies, and names hidden if you want them hidden. Inside a team, a deliberation: each member picks an option and writes why before seeing anyone else’s, everything is revealed at once, the member who answered first opens the discussion, and the same question comes back afterwards so a changed mind is recorded. Marked on the position and the reasoning - per student, either way.
DELIBERATION
YouCounter at a higher price
DevAccept it now
AmaraDecline and keep growing
All three locked in first · Dev opens the discussion
Group conversation
A room where the team talks, not the AI
A thread with no AI in it: the members of a group discuss, negotiate, plan or push back on each other, and the transcript is the work. Set how many messages each member must write, join in yourself when you want to, and give the room to only some members with a roles block. Turning either of these on turns group work on: every member gets their own grade, and the shared submission saves card by card as they type.
GROUP CONVERSATION
the receivable is the risk, not the margin
Amara then why did you rank marketing first?
2 of 2 messages each · no AI in the room
Boards, canvases & roles
Room to work together
Post-it walls, concept maps and wide canvases the class fills in live. Group activities where every member is handed a role that changes what they do - the analyst, the sceptic, the recorder, the presenter.
POST-IT BOARD
late payersone big clientsteel orderno credit line
AnalystScepticRecorder
Everyone posts at once · roles handed out for you
Learning-outcome pages
The page they read before they start
Not everything is an assignment. Say what students should be able to do and Conjure lays out a clean reference page - the outcomes, the supporting material, worked examples, a diagram they can pull apart. Nothing is submitted and nothing is graded; it’s the thing you’d have written by hand at 11pm. More below.
LEARNING OUTCOMES
Self-paced · nothing to submit
Students will be able todistinguish a liquidity gap from insolvency
Students will be able toread a cash-flow statement for timing risk
And the classics
Marked by a rule, in a moment
Multiple choice and select-all, short and long response, numeric answers, multi-part questions, prioritization, drag-and-sort, click-through reveals, guided journeys - every one with a model answer and a feedback note behind it.
MULTIPLE CHOICE
What is Nordvale’s actual problem?
Insolvency
A liquidity gap ✓
4 pts · model answer and feedback note behind it

The blue cards run on AI while the student works; the rest are checked the moment they submit. Every part carries its points, its model answer and the note on why it’s there - so you can read what a student did, not just what they scored.

Debate, peer review & study

Argue it. Review it.
Practice it.

Describe the debate, the review round or the study set you want, and Conjure builds it into the assignment - from your own course material, personalized to every student.

Debate & deliberation
An argument they have to defend
Students take a position and defend it - against classmates, or against an AI opponent that takes the other side, builds the strongest case for it and presses on the weakest step in their reasoning. Say who they are up against and Conjure builds it: the whole class, each committing before they see anyone else’s answer; a group of three or four; or one student against the AI. You read the reasoning, not just the vote.
AI OPPONENT
Takes the other side · presses on the weakest step
raise prices - our customers will stay
Opponent Your own churn data says the bottom tier leaves first. Which customers are you counting on?
Peer review
Draft, review, revise
Students submit a draft, give each other feedback against your rubric, then revise. The prompt can be built from your course material - the same companies, numbers and cases they are studying - so it is too specific to hand to a chatbot, and specific enough that the feedback has something to hold on to.
PEER REVIEW
Draft · two reviews · revision
MayaYour forecast skips the receivable - where does Q3 cash come from?
DevStrong opening. Tie the ratio back to the board’s question.
Study sets & practice
Practice built from your own course
Flashcards and practice quizzes generated from your slides, readings and notes, and set in each student’s own interests - so they practice the concept, not an answer key. When a Lens session ends, the lecture can become a study set too.
STUDY SET
From Lecture 9 · card 4 of 15
What does a current ratio below 1 tell you about Nordvale?
It can’t cover short-term debts from current assets ✓

In class, Conjure Live builds the same kinds of activity onto a slide in Lens - a vote, a quick check, a debate - and the whole room takes part at once. See Conjure Live in Lens

Simulations

Not a worksheet.
The job itself.

Your students are walking into the first job market where the routine work is already automated. The part that is left - reading a situation nobody briefed them on, making the call, defending it to someone who wanted the other one, directing the AI instead of being replaced by it - is the part a simulation can actually give them. Ask for one and Conjure builds it from your course in minutes.

1
They take the role
A real company with real numbers: financials, a team, a market position, and a problem waiting on their desk on day one. Analyst, founder, GM, clinical lead - the job they are two years from applying for.
2
The calls cost something
Allocate the budget wrong and they live with it in the next stage. Get the analysis right and the story pushes them further. Nothing is checkbox-style: every choice reshapes what comes next, and a career ladder can gate on whether they earned the promotion.
3
A mentor who knows where they are
The AI mentor knows the stage, the decisions made so far and where the reasoning broke - and pushes back on that, not with a canned hint. Ask it a question and it answers inside the scene.
4
They defend it to someone who pushes back
A memo, a board recommendation, an email thread with the CFO who wants the number explained, a client who has already decided you are wrong. The work of the job, marked on whether the reasoning holds - not on keywords.
5
You see everything
Stage completion, where the class split, which decision point tripped the most students, what the mentor kept flagging - in your dashboard before the debrief.
Built from your own materials, set in a world each student recognises, and run as take-home work here or live in class through Lens.
See a full simulation
AI learning, and learning to use AI

Not pasted into a chatbot.
Directing one.

Every copy is personalized and every conversation is live, so there’s nothing to hand to a chatbot. Instead the AI is inside the work - and the skill being practised is using it well. More on the teaching side of that: teaching students to work with AI.

Directing an agent is the assignment
Give students a data assistant that explains but won’t solve, a Coding Agent that builds what they specify, a page they have to describe precisely before it exists. The transcript shows whether they named the problem, specified the fix, and checked the result - and that is what gets marked.
Feedback while it can still change something
A wrong answer says what the wrong answer implies they believed, not just that it was wrong. Written reasoning comes back with feedback the moment they submit. The mentor inside a simulation pushes on their specific thinking, at the stage they are on.
A mentor beside the work
With Class Agents & Workspace on, a student stuck on Part 3 asks the Mentor and gets the brief quoted, or asks the Q&A Agent what’s due and gets the answer with its source. Every published activity is something the agents can cite - and you see every conversation.
AI-resistant by design
Forty-two students, forty-two versions. A character who answers what they actually said. A build that only works if the request was precise. The shortcut stops paying, and the reasoning is graded either way.
The same principle runs through everything Kova builds: AI in the moment, responding to what each student actually does.
See AI-assisted learning
Learning-outcome pages

Not everything
is an assignment.

Some of what students need is just a good page to read - what they should be able to do by the end, the material that gets them there, and a worked example or two. Conjure builds that the same way it builds the graded work: say the outcomes in a sentence, get back a clean page. Nothing is submitted, nothing is scored.

FIN 301 · Week 4Self-paced · nothing to submit
Liquidity, solvency and the gap between them
Before Thursday’s case. Read this once, then try the two examples at the bottom - they take about ten minutes together.
By the end of this week you will be able to
01Distinguish a liquidity problem from an insolvency problem, and say which one a given firm has.
02Read a cash-flow statement for timing risk rather than for the bottom line.
03Name two instruments that bridge a liquidity gap without adding leverage, and the cost of each.
Worked example
A profitable distributor with 90-day receivables and 30-day payables. Solvent on paper, out of cash in March.
Where the gap opens
Cash out by month · March is the squeeze
Interactive, not a handout
It’s built from the same parts as everything else, so the page can carry a chart students change, a diagram they pull apart, a slider that shows what happens to the outcome when one input moves. A static PDF makes them read about the idea; this lets them push on it.
Written as outcomes, not topics
Every line comes back as something a student will be able to do, because that’s what makes a page useful to read and an assessment possible to write. “Cash-flow statements” is a topic. “Read a cash-flow statement for timing risk” is an outcome, and it tells the student what to look for.
Give it what you already have
Point it at your slides, your syllabus outcomes, a reading, a chapter you’ve marked up - it builds the page from your material and your framing. Give it almost nothing and you still get a complete, professional page to edit rather than a blank one to fill.
It feeds the rest of the course
A published outcomes page is something Class Agents can cite, so a student asking the Q&A Agent what Thursday covers gets the answer with its source. And the outcomes you wrote here are the ones the graded work is built against.
Outcomes pages sit alongside three other things Conjure builds from the same sentence: graded activities, games, and self-paced study aids.
See the full vocabulary
Personalized, per student

Forty-two students.
Forty-two versions.

The concept you're testing stays fixed. The world it's set in doesn't. Conjure re-skins every student's copy to something they actually care about — so no two submissions are the same, and copying a neighbor (or a chatbot) stops being useful.

The concept · fixed for everyone
Given a firm that's profitable on paper but short on cash, identify whether it faces a liquidity or a solvency problem, and explain what the current ratio does — and doesn't — tell you.
The context · one per student
MayaMusic
A record label clearing royalties it can't yet spend…
DevonBasketball
A sports franchise with a sold-out season but a cash crunch…
PriyaPublic health
A clinic network booked solid but late on supplier payments…
Because every copy is grounded in a different real-world context, the work resists the copy-paste-into-a-chatbot shortcut — students have to reason it through in their own setting.
See AI-resistant assignments
Grading

It grades the thinking,
not just the answer.

Conjure grades against the rubric you'd use anyway — checking whether the reasoning holds, not just whether the final number matches. It drafts feedback in plain language, scores every submission, and surfaces the handful that actually need a human. You keep the final say.

Devon R. · Q2 · submitted 11:38pm
9/10
Student answer
It's a liquidity problem, not solvency — the franchise is profitable over the season, but ticket revenue lands after payroll is due. The current ratio looks healthy because season-ticket receivables count as current assets, but they can't cover this month's cash needs.
✓Correctly distinguishes liquidity from solvency
✓Identifies the timing mismatch as the core issue
✓Names the current-ratio blind spot (receivables ≠ cash)
Feedback drafted for you
Strong reasoning on the timing mismatch. To reach full marks, quantify the shortfall — roughly how much cash is needed before receivables clear? One number would sharpen the argument.
✓ Score & feedback synced to Canvas
The other half of Conjure
In class, Conjure builds live activities inside Lens the moment you need them - a poll, a sim, a debate, pushed to the room in seconds. For take-home, Conjure Assign builds the coursework and grades it. One builder, two jobs: the room in front of you, and the week ahead. And everything it publishes is a source the Class Agents can quote when a student asks.
See Conjure in Lens →

Put your next assignment
on autopilot.

Bring a syllabus or a single reading. We'll turn one unit of it into a personalized, auto-graded assignment while you watch.

Request a demo