Most education software talks about features. The clearer question — borrowed from Clayton Christensen — is: what jobs are faculty hiring a tool to do? Eleven jobs make up teaching a course in 2026. Six have always existed. Five are new since 2023, because students now arrive with an AI tutor in their pockets. Canvas and IgniteAI hire for the traditional jobs. Crucibl is built for the five AI-era jobs they're structurally locked out of doing.
Every faculty member running a course in 2026 has to hire for these eleven jobs. The job either gets done — by the faculty member, by a tool, or by some combination — or the course quietly degrades.
The five AI-era jobs are why faculty are exhausted. They are real jobs faculty are now being asked to do. They have no historical playbook. Until very recently, no tool was built to hire for them. The cat-and-mouse era — detection tools, integrity policies, AI-use disclosures — was the first attempt to address them, and it has failed comprehensively.
Each of the five AI-era jobs traces to a documented failure mode in the current AI-in-education record. Together they define why Crucibl exists.
The instinctive response to AI in higher education has been detection — Turnitin's AI checker, GPTZero, integrity policies. Detection has failed. The tools produce false positives that disproportionately punish non-native English speakers and false negatives that miss lightly edited output. Even a perfect detector would tell you only that a student used AI. It would tell you nothing about whether the student learned. The job is to stop the arms race and design AI use into the assignment.
A 2026 systemwide survey of 94,000+ California State University respondents found that 67% of students say their professors don't teach them how to use AI effectively. That's not a faculty failing — it's a tooling failing. There has been no method, no curriculum, and no software that supported faculty in actively teaching prompt crafting, verification, and ethical use. Students were left to figure it out themselves, mostly badly.
Cognitive Load Theory distinguishes intrinsic load (the irreducible difficulty of the material), extraneous load (load imposed by poor design), and germane load (the productive struggle that builds long-term schemas). Lepine et al. (2026) showed that unstructured AI introduces approximately 3× more extraneous load than the underlying task. Bastani et al. (2024) showed unconstrained GPT-4 access produced 17% worse independent performance once AI was removed. The job is to engineer the learning so AI augments rather than displaces the productive struggle.
When the dean asks "did your students actually learn?" — have a real answer. When a tenure committee asks "what's the evidence?" — have it. When an accreditor asks "how is your AI strategy aligned with your learning outcomes?" — produce documentation that holds up. Faculty don't need notes for themselves; they need artifacts that survive curriculum committee, accreditation review, and tenure-and-promotion files.
This isn't a feature you can ship. It's the felt sense of trust a faculty member has that their AI tools aren't quietly hollowing out their courses. It is earned through constraint, transparency, and audit — and lost through detection theater and unstructured chatbot adoption. The job is real even though it sounds soft. Faculty who don't have it leave the profession.
Three tools, three different scopes. Steel-manned honestly: Canvas does what Canvas does well. IgniteAI extends Canvas into productivity assistance. Neither hires for the five AI-era jobs.
Canvas is the system of record for your course. Hosts your modules, your gradebook, your discussion forums, your file uploads, your administrative envelope. About 40% of US higher education runs on it. That's not a weakness — it's a deliberate scope choice. Canvas was built to be the system of record, not the system of pedagogy.
Hires for jobs 3, 5, 6.
IgniteAI is Canvas's productivity layer. Generates rubrics on demand, summarizes discussion threads, drafts image alt text, reorganizes course materials. If you're using IgniteAI today, you're getting genuine productivity value: less time on administration, more time with students. That's a real win, available out of the box.
Hires partially for jobs 1, 2, 4 plus faster 6.
A research-grounded 10-Principle Framework, a multi-agent pedagogy stack — Builder, Socratic Tutor, Critic-Coach, Instructor Insight Agent — and a patent-pending audit-trail infrastructure. Crucibl runs alongside Canvas as a Canvas LTI Advantage tool. Faculty walk out with four artifacts: Course Architecture Document, Constraint Set Library, Audit Trail Report, Outcome Package.
Hires for jobs 7, 8, 9, 10, 11 — and a deeper version of 1, 2, 4 specifically for AI-mediated work.
The five AI-era jobs are uncovered by Canvas + IgniteAI. Not partially. Uncovered. IgniteAI is a productivity layer; it generates content. It doesn't enforce constraints on student-AI interaction, doesn't produce an audit trail of process, doesn't teach AI literacy, doesn't manage cognitive load, and doesn't produce documentation defensible at external review. Those aren't missing features that could be added in a sprint — they require a different data model and a different tool philosophy.
We get this question a lot. The honest answer is that Canvas could add it — but the path is acquisition, not internal build. Two structural reasons.
Canvas's data model is courses, modules, assignments, submissions, and grades. Those represent the outputs of teaching. The data model required to run AI-mediated learning is different: sessions, AI personas, constraint sets, Socratic checks, deliverables, audit trails, persona-bound interaction logs. The two models aren't compatible. Bolting one onto the other isn't a small feature add — it's a multi-year rebuild that breaks every existing customer integration.
Canvas sells to institutions through a procurement cycle measured in months. The institutional buyer asks "does this LMS run my courses?" — not "does this tool help my faculty deliver AI-augmented learning that doesn't undermine the discipline?" The faculty pain on jobs 7–11 doesn't show up in the institutional RFP. Even if Canvas could rebuild the data model, the economics don't justify it.
The economical path for Canvas to add Crucibl-class capability is to acquire a runtime that runs alongside Canvas — the same way Canvas already hands off to ProctorU for proctoring, MasteryConnect for assessment (Instructure acquired it in 2019), Portfolium for credentialing (acquired 2019), or LearnPlatform for analytics (acquired 2024). Instructure's pattern is to buy specialized runtimes that fit alongside Canvas, not to rebuild Canvas's data model.
The realistic future is either "Canvas acquires a runtime like Crucibl" or "your institution adopts a runtime like Crucibl now and has the evidence by the time the institutional AI-strategy conversation reaches your dean's desk."
Microsoft bundled Copilot into 365 at $30 per user per month for two years. Gartner ROI studies show enterprise customers cannot justify the spend. Forrester named "capability without workflow" the top failure mode. Microsoft itself has publicly experimented with non-OpenAI models because the current state isn't working.
Free AI bundled into your LMS will follow the same arc. Faculty bolting Canvas's bundled AI onto unchanged course designs replicate the Copilot experience: AI features turned on, outcomes unchanged, accreditors unimpressed. The methodology layer is what makes the capability useful.
Calculators were once banned in math classrooms. Spellcheckers were once derided as crutches that would destroy student writing. Both were eventually normalized — not by faculty surrendering on rigor, but by faculty redesigning what they were teaching and how they were assessing it. The AI arc is the same, accelerated. The work is concrete: redesign assignments, clarify permitted use, explicitly teach prompt crafting, verification, and ethical use. That redesign is the work. Crucibl operationalizes the work.
You need two things to deliver an AI-augmented course in 2026: a system of record (Canvas, almost certainly) and a runtime that hires for the five AI-era jobs. Canvas + IgniteAI is the system of record plus productivity assistance. Crucibl is the runtime that handles the five AI-era jobs Canvas + IgniteAI doesn't touch.
Canvas runs your course. Crucibl runs the part of your course that involves AI.
Crucibl ships as a Canvas LTI Advantage integration. Your faculty stay in their Canvas workflow. Your students stay in their Canvas workflow. Crucibl is the engine running underneath the AI-mediated parts. Everything else stays in Canvas.
A field guide for faculty, department chairs, deans, foundation funders, and grant reviewers. Each principle prevents a documented failure mode in the AI-in-education research record. Together they define what hires for the five AI-era jobs.