The discipline

The student can articulate why ASE is a discipline shift rather than a tooling shift, situate the current transition within the historical arc from ENIAC patch cables to vibe coding, name the empirical evidence for the speed-versus-trust gap (68% PR delay rate, solve-rate collapse under audit), and apply the autonomy-level framework to any agentic tool.

The discipline

The student can sort developer work into the three categories (eroding, stable, compounding) with concrete examples, recognise the four structural risks of the discipline (skill erosion, black-box codebases, responsibility gaps, training-data bias), articulate Polanyi’s residue as the conceptual ground for the compounding category, and name the four barriers to industrial adoption identified in the 2026 systematic review.

The discipline

The student can use the practitioner mental model of agents (context window as the agent’s whole world, tools as action space, planning as next-step generation under context) to predict where a given prompt will succeed or fail, and can diagnose agent failures against the four-mode taxonomy (hallucination, misalignment, ambiguity collapse, sycophancy).

The discipline

The student can decompose any agentic workflow into its seven parts (intent, specification, context, plan, execution, verification, audit trail), explain the auditability principle and its operational consequences, and recognise an audit trail produced as a byproduct of agentic work.

The discipline

The student can read any ADE as an instance of the six-pillar architecture, classify it into one of the three ADE categories (terminal/CLI, IDE-integrated, browser-based), evaluate a permission configuration in terms of blast radius, and apply the minimal-footprint principle to a concrete sandboxing decision.

The environment

The student can produce a framing document for an agentic project (problem statement, stakeholder list, testable definition of done, out-of-scope list with defensible boundaries), apply Brooks’s four-way taxonomy to a constraint set, and run a reverse-interview with the agent across the six question clusters (purpose, user, success, boundaries, constraints, testability).

The environment

The student can describe a modern web application architecture at the level needed to direct an agent intelligently, trace a request through the full cycle from button click to database query and response, and recognise when an agent’s proposed architecture is sound, broken, or merely fashionable.

The environment

The student can specify a user interface at directing-quality precision (user flow, information hierarchy, interaction model, feedback design), distinguish descriptive from explanatory documentation, and recognise illegibility in an interface, codebase, or document as a control problem rather than an aesthetic one.

The environment

The student can distinguish using AI to build software from building software that contains AI, address the architectural questions cognification raises (inference location, latency budget, model choice, failure handling), and apply agent-economics reasoning (token pricing, prompt caching, frontier-versus-local trade-off) to a product design decision.

The engineering

The student can write a complete specification containing goal and business rationale, verifiable success criteria, architectural guidance, validation approach, and known pitfalls; can diagnose categories of agent failure against categories of specification gap; can run the reverse-interview methodology to extract specification material from underspecified intent; and can articulate the rationalism-empiricism axis and ASE’s empiricist position, recognise the co-evolution of intent and build, execute the mechanics of one turn of the spiral (intent revision, context update, safety positioning, build, verification, review), make commit-point decisions, and detect drift as it develops.

The engineering

The student can design a project-specific CLAUDE.md or AGENTS.md file with appropriate content across the four kinds (standards and processes, code quality expectations, problem-solving heuristics, collaboration protocols), maintain a lessons-learned file across tasks, and recognise context degradation in its three forms (overflowed budget, ad-hoc artefact, neglected maintenance).

The engineering

The student can apply Git practices appropriate to agentic work (pre-invocation commits, atomic commits with rationale, branch discipline for safe agent output evaluation), configure permission boundaries that scope blast radius, and execute trajectory-management interventions in response to the standard drift signals during a session.

The engineering

The student can design verification gates appropriate to the failure modes a project actually faces, recognise verification theatre when they see it, resist gate bypass under deadline pressure, and produce evidence sufficient for the MRP standard.

The engineering

The student can distinguish implicit from explicit orchestration, decide when the cost of explicit orchestration is justified (given the 15x token premium), break a task into sub-tasks with clean boundaries and isolated context, and apply each of the five workflow patterns (prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer) to a concrete decomposition.

The engineering

The student can produce a coordination design for a multi-agent workflow that specifies patterns, roles, boundaries, and cost-benefit justification, can teach an orchestrator to delegate effectively, can scale effort to complexity, and can design tools for the agents to use.

The engineering

The student can review foreign code (agent-generated or legacy-inherited) against the MRP’s five criteria, distinguish surface correctness from rigorous-audit correctness, apply the dependency-graph methodology to a legacy migration, and rewrite tests from logical intent rather than literal translation.

The engineering

The student can identify prompt-injection vectors in an agentic system, design permission boundaries that contain blast radius, evaluate an agent’s tool-server dependencies for malicious servers and supply-chain risk, apply the OWASP Top 10 for Agentic Applications as a practical reference, and distinguish security questions (what could happen) from governance questions (what is sanctioned, recorded, and reportable).

The market

The student can analyse a startup case study for value-creation mechanism, organisational enabling condition, and managed-or-ignored risk; can apply the two-mechanism frame (cost compression and cognification) to a new opportunity; and can name the new bottlenecks that replace development cost and timeline as primary obstacles.

The market

The student can articulate the AI-affects-tasks-not-jobs thesis with empirical support, identify the human-sandwich pattern in real workflows, read employer signals that distinguish genuine ASE practice from performative mention, and evaluate a role for whether it will develop or erode the compounding skills.