A technician is partway through a repair, with both hands occupied, when they need to check an instruction. Getting that information means putting down their tools, finding the relevant section of a manual, then applying it to the equipment in front of them. If the correct action is still unclear, they have to go searching for a video or call someone who knows.

Each interruption sets off the same back-and-forth, with the worker bouncing between the task at hand and the scattered resources needed to work through it.

Making a technical procedure conversational means the guidance comes to the worker, instead of the worker going to find it. An AI assistant for physical work leads them through approved instructions verbally and visually, with room for questions along the way.

The assistant guides the work one step at a time

The assistant leads the walkthrough the way a human expert would. It gives the worker an instruction, shows them how it’s done, and answers any questions they have.

Let’s say a technician is replacing a control module. After telling them to remove the access cover, the assistant pauses while they do it. If they need to know why the step matters, or they run into an issue removing it, they can ask as many questions as they need until they’re ready for the next step.

They can also go back by simply asking to revisit an earlier step. The assistant takes them there, then returns them to where they were in the procedure — no hunting through a manual to find their place. So they can take the time to understand a detail without losing the thread of the work.

The same holds true whether a procedure has five steps or a hundred steps with branching logic. It's still one conversation, and the assistant keeps it moving.

The conversation is visual as well as verbal

Written instructions only go so far. A part name or number can be exact and still leave someone hunting for it on the equipment. An instruction needs to show as well as tell.

Researchers have long found that people learn less when they have to mentally piece together instructions and diagrams presented separately, a problem known as the split-attention effect. Bringing the explanation and the relevant visual together helps reduce that extra mental work.

Most companies already have the materials for this: drawings, photos, videos, and the engineering models themselves.

As the source of truth for a product, a 3D CAD model provides a detailed visual reference for its parts and how they fit together. It's also the hardest to get in front of a worker. Most frontline tools can't easily work with 3D, which has often led to building custom applications.

The payoff is what a CAD model can do that a drawing can't. Every part is its own object, with its own identity and place in the assembly. That's what lets a procedure step attach to the exact component it means. The assistant can highlight the specific part an instruction refers to. The worker can point at a part on the model and ask what it is, even without knowing its name.

Whether the assistant is pulling up a drawing or video, or manipulating a 3D model, the visuals are part of the guidance, not a separate resource the worker has to go find.

The procedure stays the same, even as the conversation adapts

Two technicians can have completely different conversations with the same procedure. Someone learning the task may want to understand why a step matters. An experienced technician might only need to confirm a specific detail. What changes is the explanation, not the underlying procedure.

That's deliberate. In physical work, a skipped step or the wrong part can stop a line or get someone hurt.

So the approved procedure is the source of truth for the steps, the precautions, and any branches through the work. When a step depends on the worker's answer, the assistant follows the path that’s defined by the procedure. It can explain, repeat, and answer questions without adding or skipping steps.

That procedure isn't invented by AI. The starting point is the company’s own material: existing procedures, manuals, engineering models, and technical references. The AI assistant turns that material into step-by-step guidance tied to the relevant parts and visuals. Experts can preview the experience, refine it through conversation, and approve it before workers use it.

Questions are answered from the same sources, with references people can check. When those sources don't cover the situation, the assistant says so instead of guessing. That matters most when something unexpected happens: the limits of the guidance need to be as clear as the instructions themselves.

Staying in the work

Spark is Campfire's AI assistant for physical work. It turns technical procedures into conversations so workers can stay in the work and get guidance as they go.

The same guidance can be adapted for someone learning a procedure, someone carrying it out on the job, or someone troubleshooting an issue.

For instance, a technician on a service call can set their phone beside the equipment and talk to Spark using AI glasses, earbuds, or just the phone’s mic. Spark talks them through the steps, bringing up the relevant model view or video on screen when it’s needed. They can glance at the visual while keeping their hands on the task.

If you’re interested in bringing Spark to your team, explore the Spark Preview to see how it could work with one of your own procedures.