AI as a Copilot Keeps Engineers in Control of Electronics Design

AI & Compute

AI as Copilot: Why Engineers Keep Control of Electronics Design

AI is entering electronics design as a copilot, not an autopilot: agentic workflows from firms like CELUS automate routine work while engineers keep design responsibility and verification control.

By
Tom Whitfield
Filed
Channel
AI & Compute
Read
4 min read

Electronics design does not reduce to producing an answer from a set of inputs, and that is precisely why AI is entering the field as a copilot rather than an autopilot. The emerging model pairs a system that can investigate options, perform repetitive work and flag problems with an engineer who keeps design ownership. CELUS, whose CEO Tobias Pohl co-authored the argument, is building its Design Assistant around exactly that split.

From Automation to Assistance

Engineers already drown in information. Component datasheets can run to hundreds of pages, and selecting a device means weighing specifications, availability, cost, power consumption and compatibility with other parts. The bottleneck is not a lack of tools; it is the volume of information and the number of possible pathways that engineers must navigate within finite project schedules.

The problem is sharpest at the start of a project. Requirements are incomplete and architecture is unsettled. Engineers work in what practitioners call "fuzzy" space: they know what the product must achieve, but not the optimal implementation strategy. AI helps here by turning loosely defined intent into something structured. An engineer can describe a function in natural language or sketch it roughly, and the system identifies required building blocks, asks about missing information and presents possible approaches. The "what" stays with the engineer.

What Does the Separation of 'What' and 'How' Achieve?

A useful principle is to separate what the system must do from how it is implemented. High-level functional architecture provides that separation. Instead of committing immediately to components and connections, the engineer defines major functional blocks and their relationships, creating a representation of design intent that supports later refinement.

At this stage AI can search component information, identify candidate devices, draw up functional circuits and generate an initial schematic or bill of materials. It also makes it practical to explore several implementations rather than spending significant time developing just one. The engineer can still ask why a particular approach was proposed, change constraints, reject a component or choose a different architecture. Routine decisions can be automated; decisions that require product-specific judgment stay human.

Why Agentic AI Changes Matter

Early AI-assisted tools behaved like a conversational interface bolted onto a large language model: the engineer issued a prompt and received a response. Agentic systems take a more structured route. An agent breaks a goal into tasks, executes them in sequence with specialist capabilities, and checks the result against the original demands.

That capability fits electronics design well. A robust workflow begins by capturing requirements and separating what is known from what has been assumed or left unspecified. It then creates a plan, executes individual stages, and verifies the resulting design. If a requirement changes, the agent revisits the process rather than modifying one isolated output. This behavior is easier to supervise: a visible plan, progress through defined stages and explicit verification give more insight than an opaque system handing over a finished design and expecting trust.

Why Verification Cannot Be Optional

The physical nature of electronics raises the verification burden. A software developer can run a test suite against a change; electronics requires its own layers of checking, from electrical connectivity and design rules through simulation, signal integrity, power integrity, manufacturability and other physical constraints. As AI agents gain greater ability to modify design data, these checks become more pertinent.

One promising approach makes verification continuous rather than final-stage. An AI-generated component choice could be checked against the requirements that led to its selection; a proposed change could trigger design-rule or simulation checks, with results returned to the engineer or used by another agent to refine the work. The principle is straightforward: AI should generate, and also demonstrate that what it generates stays within engineer-established boundaries.

The Engineer as Orchestrator

Engineers should not become supervisors of disconnected AI systems. If information must be transferred manually between component-selection software, schematic capture, simulation and mechanical design, adding AI only creates another layer of complexity. The more useful vision has agents and tools exchanging information while the engineer retains control of overall design intent. Connected agents could eventually flag how an electrical change affects thermal performance, mechanical packaging or power requirements, making cross-domain engineering practical. The engineer becomes an orchestrator rather than an operator of individual tools — greater leverage for expertise, not a reduction of it.

The strongest argument for AI in electronics design is that it makes engineering judgment more valuable, not redundant. As Pohl puts it, the AI copilot processes information, explores routes and flags issues ahead — "however, the engineer still decides where the aircraft is going." CELUS is evolving its Design Assistant toward an agentic architecture on that basis, keeping decisions open to the engineer rather than presenting AI results as immutable answers. If the workflow respects the line between automating work and automating responsibility, engineers will not have to choose between embracing AI and staying in control.

Original: celus.io

Share this article:

More from Tom Whitfield

Tom Whitfield

Show full bio

Staff writer covering companies, deals and strategy in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,054 articles

Related articles

« Previous articleNext article »