An agentic engineer, as defined in Trustworthy AI Software Engineers, is an AI agent that satisfies four conditions before it can be considered an AI software engineer rather than merely an AI coding agent. The definition is deliberately set higher than “autonomous coding”: software engineering is treated as extending well beyond programming into requirements elicitation, specification analysis, architectural design, testing, and long-term maintenance, so an agent must be judged against that fuller scope of activity.
Usage
The four conditions are: (i) the agent must be able to handle a range of software engineering tasks beyond coding, such as evolving problem statements, requirements, constraints, ethics, design, testing, and operations, under iterative refinement and partial information; (ii) the agent must demonstrate agency through planning and tool use, extending beyond code-centric tasks; (iii) the agent must be collaborative, integrating into human workflows by supporting coordination, feedback, and negotiation, since software engineering is fundamentally team-based; and (iv) the agent must respect human values, constraints, and ethical responsibility, justifying its decisions and supporting accountability and oversight.
Trustworthy AI Software Engineers frames this as a shift in the role of human engineers, from directly performing tasks toward supervision, orchestration, validation, and ethical oversight of agentic engineers operating within human-AI software engineering teams, rather than a model of AI replacing human engineers.
A second and incompatible use of the same term comes from From Determinism to Delegation: AI-Native Software Engineering and the Evolution of the Agentic Engineer, where the agentic engineer is a human professional archetype rather than an AI agent: the practitioner of AI-Native Software Engineering, whose primary artifact is the agentic system rather than the program. That paper’s compressed statement of the distinction is that software engineering builds the system while agentic engineering builds the agentic system that helps build, operate and evolve the system — software engineers optimised for deterministic design and implementation, agentic engineers for orchestrating probabilistic collaborators safely and productively. Readers should note that the two sources use “agentic engineer” for opposite sides of the human-AI relationship, and neither acknowledges the other.
That paper characterises the human archetype across fourteen dimensions, and along three axes it treats as the paradigm shift itself: the unit of work moves from code implementing a feature to a supervised agent workflow; the correctness model from binary assertion to statistical evaluation under uncertainty; and accountability from authorship to outcome ownership. It names the scarce skill as judgment — writing a specification precise enough for an agent to execute, then detecting the plausible-but-wrong output a deterministic test would never flag — and is explicit that its fourteen-dimension table is an analytical scaffold rather than an empirically validated taxonomy.
On what the role demands, that paper maps agentic-engineering practice onto the SFIA 9 responsibility ladder, from executing deterministic commands via generative assistants at Level 1, through authoring asynchronous integration code and configuring vector databases, to constructing multi-agent state loops and evaluation pipelines at mid-levels, and leading ISO/IEC 42001 and NIST AI RMF audits and governing model portfolios at the top. It decomposes competency into technical skills, soft skills, knowledge and abilities, and reports that educators increasingly argue curricula should prioritise problem definition, system design and debugging/evaluation over code authoring — a stance it notes is consistent with its own evidence that value migrates toward oversight. It marks one common claim as unsettled: that senior engineers adapt more readily is, it states, a hypothesis rather than a settled finding, and the controlled evidence on which cohorts benefit is mixed — a lab experiment and three field experiments report substantial gains, largest for less experienced developers, while a randomized controlled trial of 16 experienced open-source developers on their own mature repositories found a 19% increase in completion time.
Related Terms
Trustworthy AI Software Engineers, Trustworthiness Dimensions, AI Coding Agent, Structured Agentic Software Engineering (SASE), CRAFT Values and Principles for Agentic SE, Vocabulary Design Considerations for Agentic SE
- From Determinism to Delegation: AI-Native Software Engineering and the Evolution of the Agentic Engineer — source of the second, human-archetype account above
- AI-Native Software Engineering — the paradigm that account says produces the role
- Supervised Agency Spectrum — how the oversight it centres on is graduated
- Compositional Reliability — one of the structural limits it says the role must work within
Not to be confused with Agentic Engineering, a distinct term for the discipline or practice rather than a definition of the agent or the person; that page carries its own competing accounts.