~/rewind/copia/agents

Copia (opens in a new tab)Case study · 2023 – 2024 · 7 min read

One agent, many sites.

How a question I kept asking in my first months, "why does an agent have to live at one site?", became multi-site agents, and changed how Copia's customers pay for backups. A story about asking "why not?", and about treating the business model as design material.

Screens and names use dummy data, to protect Copia and its customers.

★ The futureAn agents page that shows each agent's day like a timeline, so you catch failures before they spread
⏪ Rewound toOne agent at the organization level, picked from a dropdown in any project
▶ Built forwardDashboards for agent use and status, with room to grow into bulk changes and offline backups
My role
Senior product designer, and Copia's only designer at the time
When
My first three months (late 2023) through launch in Dec 2024
Humans
My PM and I drove it, with engineering, finance and brand in the room
My part
Asking why, building the case, the design of agent choice, permissions, icons and status, and driving the business questions behind pricing
What changed
Multi-site agents shipped Dec 2024 ↗. Existing accounts grew, and new customers who had held back over cost signed on.
1agent, any number of sites
1×each vendor license, not once per site
0new roles: built on existing permissions

01Meet DeviceLink

Backups for the code that runs a factory

DeviceLink backs up the code on factory machines. Every site needed its own agent: a small app on a machine in that building.

Copia started as Git for industrial automation code. Then it grew into backups: DeviceLink pulls the code off PLCs on the floor and keeps every version.

No one else backed up this many PLC vendors automatically, so DeviceLink was the big bet. The model was simple: an organization has sites, and sites have devices. To back up a site, customers installed an agent, a small app on a machine in that facility. The agent talks to the PLCs on one side and to Copia on the other, and uses each vendor's own software to read their files.

Before: one agent per site, and each agent could only see its own site.

Copia's team of technical support engineers (TSEs) set agents up with each customer's controls engineers. Being on-prem had a real upside: the agent lived inside the customer's walls, under their control.

02The cost of one site

Five problems, one root

The more I learned, the more the same root kept showing up: the agent was tied to a building.

Availability

Agents went offline

The laptop died, went to sleep, or someone shut it down. Backups quietly stopped.

Money

Licenses stacked up

Each agent needed the vendor software for every device it backed up, paid monthly. Across many sites, that added up fast.

Money

Idle agents still cost

A site with three devices still needed a whole agent: its own machine, and its own copy of every vendor license.

Capacity

One backup at a time

An agent runs one job at a time, so each site had a ceiling on how much it could back up in a day.

Availability

A machine, always on

Every site had to keep a computer running just for the agent.

The biggest gap

Sites change over time

While a site is being commissioned, it may need backups several times a day. Once it's running, far fewer. One fixed agent per site was either swamped or idle.

Illustrative, not real data: a site's backup load across its life. Swamped, then mostly idle.

03Listening first

"Why can't the agent live in the cloud?"

New to the industry, I asked a lot of questions, and kept asking “why not?” about agents tied to one site.

I came to Copia without an industrial background, so in onboarding I asked a lot of questions. Engineers are always happy to share an idea or an "I wish".

  • Engineersand engineering managers
  • Sales & CSdeals and escalations
  • Supporttickets and TSEs
  • Recordingsendless customer calls

The idea of agents that weren't tied to a site was already floating around in those conversations. It wasn't mine alone. What I did was keep asking "why not?" until someone had to answer.

The pushback

The architecture was built around one agent per site, so this meant real engineering change. And no customer was asking for anything this radical. But under the "no" there was curiosity, and that was enough to keep going.

04Building the case

From a hallway question to the roadmap

My PM and I took the question to engineering, finance and brand, until it made the roadmap.

Over the next few months, my PM and I took the question to anyone who could tell us what would break.

He brought it to the engineering managers. Together we pulled in finance and brand, to weigh the pros and cons at the idea stage: not only what it would take to build, but what it would do to the business.

  1. OnboardingAsk everyone everything
  2. Months 1–3"Why not?" in hallways and 1:1s
  3. With my PMEngineering, finance, brand
  4. Turning pointOne cross-functional review
  5. ShippedMulti-site agents, Dec 2024
The turning point

A meeting where business, money, UX and engineering looked at the idea together. Once everyone could see the whole picture at once, it went on the roadmap, and I led the design.

05The pricing debate

Fewer agents, and still a better business

If customers could share agents, would they buy fewer? I started those conversations, and my questions helped the team make the call.

The hardest conversation wasn't about pixels. It was about money: if customers could share agents across sites, would they buy fewer of them?

Single-site agents

At least one agent per site. Every agent needs a license for every vendor at its site, so the same license is paid for again and again.

Multi-site agents

Pay for the agents you actually run. Give one agent Siemens, another Rockwell, another CODESYS, and each license is paid for once.

I started those conversations and kept them going, with questions like:

  • Would this cause problems with what current customers already pay?
  • If it cuts their vendor license burden, what does Copia actually lose?

The final call was a team decision. I'm proud my questions helped us get there: an agent that fits the way a customer's sites actually work makes Copia easier to say yes to, and easier to grow with.

Who it worked for

Small orgs

Fewer agents than sites

A handful of sites with a few devices each no longer need an agent apiece.

Lopsided orgs

Agents follow the load

One plant with hundreds of devices, three with a dozen: agents go where the work is, not where the walls are.

Large orgs

One agent per vendor

Agents organized by vendor, each carrying one license, with permissions set up around those agents.

Sites in flux

Room for commissioning

Lend capacity to a site while it's being built, and take it back once it's running.

06Designing the details

Small surfaces, big questions

On screen the change looks small: one dropdown, a globe icon, and permissions that never overshare.

On screen, the change looks small. Underneath it are questions about trust, privacy and control.

1

Choosing an agent

Every project gets an agent dropdown that lists the site's own agents and the organization's multi-site agents together. Each one shows its type and a status chip: setting up, active, inactive or offline. So you never pick an agent that can't run. The project editor needed a new spot for it.

2

A globe for "many sites"

I kept it simple: site agents keep the icon people already knew, and multi-site agents get a globe. One new symbol to learn, not a new system.

3

Sharing without oversharing

A multi-site agent can back up sites that some users aren't allowed to see. Those users see only how busy the agent is, as a percentage, never which other sites or projects it's working on. Enough to choose well, without leaking anything.

4

Who's in charge

No new roles. Owners create multi-site agents in organization settings, and anyone with write permission on a project can change its agent. That matches how teams already worked.

5

A choice, not a migration

Site agents stayed. Customers who want everything on-prem keep that control. Moving is project by project: set up a new agent, then switch projects to it. The agents table was designed to grow into bulk changes.

The project's agent dropdown in resting, hover and focus states. Open, it lists site agents with a building icon and multi-site agents with a globe icon, each with a status chip: Active, Inactive, Unlinked or Setting Up.
Two kinds of agent in one list: a building for site agents, a globe for multi-site agents, and status on every row.
See the multi-site agent card and the agents table
A multi-site agent card named Dan-Agent, listing several sites, its host name and its Rockwell Logix Designer integration.
One agent, many sites, with the full list a hover away.
A table of agents with status pills such as Active, Unlinked and Setting Up, host names, keys and vendor integration icons.
The agents table, with status at a glance.

07Try it yourself

Share the agents

A made-up company with four sites. Switch the agent type, move a site from commissioning to operating, and see the agents through a plant engineer's eyes.

    A playful sketch, not the real product or Copia's pricing. Sites, vendors and numbers are made up.

    08What changed

    Customers used it in ways we hoped, and some we didn't plan

    • Projects moved over. Many existing customers moved projects to multi-site agents to cut vendor licenses and keep agents busy.
    • Accounts grew. Several existing customers expanded their use of Copia.
    • New customers said yes. Teams who had held back because all those per-site costs added up signed on.
    • New setups appeared. Large customers organized agents by vendor and set permissions around them.
    Built forward

    A few months later, we built dashboards that show agent use and status at a glance, the first step toward the agents page below.

    A site dashboard with project backup health, a job run status chart for the last 7 days, and an Agents panel listing single-site and multi-site agents with utilization bars and status.
    The site dashboard: agent use and status sit next to backup health.

    09The north star

    An agent's day, at a glance

    Where I'd like agents to go: a page that shows each agent's day, like a playlist, a calendar or a timeline.

    You'd see which sites and projects an agent backs up, when, and what failed. Then patterns would jump out: a vendor license that expired, or a bug breaking every backup for one vendor.

    Offline backups belong here too, something the industry keeps asking for: a site agent could keep backups on its own machine while offline, and sync them when it's back.

    A whiteboard concept titled visualizing device agent utilization: a grid of site-wide and multi-site agents with utilization rings, and one agent's week drawn as a timeline of backup jobs.
    Whiteboard from a working session with my PM: agents grouped by type, each with its utilization, and one agent's week as a timeline.

    ✦The future, rewound

    What this story says about how I work

    The receipts this story, in public

    1. Oct 2022DeviceLink launches for PLC backup and change detection Copia announcement ↗
    2. Dec 2024Multi-site agents are announced: one agent, any site in the organization Copia announcement ↗
    3. TodaySite-based and multi-site agents, side by side in the docs Copia docs ↗
    4. TodayPicking any multi-site agent from a project's agent dropdown Copia docs ↗
    5. Jun 2026Copia raises $26M to unify OT code management, backup and recovery Copia blog ↗

    Ask "why not?"

    Coming from outside the industry, I asked the questions insiders had stopped asking.

    The business model is design material

    The biggest design decision here was about how customers pay.

    Design for every viewer

    A shared agent had to work just as well for someone who can see one site.

    North star, small steps

    A dropdown and a globe today, then dashboards, then an agent's whole day.

    whoami // here I am

    Senior title, staff-scope work. Looking for staff product design roles.

    Complex, technical products are my favorite kind. The gnarlier, the better.

    say --hi // let's talk

    Curious how a hallway question made it onto the roadmap?