Motiv launches Opal, a platform for running your company on AI agents

One of the most interesting AI demos I’ve seen this year looks like a to-do list.

I’m sitting across from Clark Lai, CEO of Canadian venture studio Motiv, as he pulls up a workflow in Opal, the AI agent platform his company built and rolls out publicly today. He wanted to show me how quickly a workflow comes together, so that morning he’d described a job he needed done and let Opal’s built-in AI assist design one for it.

What it came back with is a sequence of steps any manager would recognize: review the source material, brainstorm options, pause for a person to make the call, then produce the finished work and have it checked.

Each step is assigned an owner. Most of the owners are AI agents (software that carries out a task on its own) but the platform lets you choose between automation or humans in the loop at specific steps. 

The plainness is deliberate. Lai hates flowcharts, arguing a simple process doesn’t need one and a complex process makes one unreadable.

“I know it’s boring, but it’s clean and it just works,” he says smiling at me.

The boring part is what has my attention. Anyone who has managed people through a process knows the fantasy was never a glowing dashboard. It’s opening your laptop to finished work you didn’t have to chase. Boring and done beats flashy and stuck every time.

He hit ‘Run’ and the workflow queued up in the background while we talked, then came back complete with a log of everything the agents did, priced out like a phone bill. This run came to about 26 cents, and the log showed exactly where each fraction of a cent went.

Lai is a member of Digital Journal’s editorial advisory committee. Because he runs a venture studio that builds companies for a living, our conversations are usually about other people’s ventures. This time he was in the interview seat, pitching his own product.

We’re in a boardroom at Motiv’s Calgary office on a warm, sunny July afternoon while the rest of the city surrenders to the Stampede. Lai sits facing me and keeps turning back to the screen to pull up another piece of Opal, which he announced in June for a group of companies piloting the platform.

What Lai gave me was a preview, and now that Opal is live we’ll be testing the workflows ourselves. What has my attention is the thinking behind the design, because it speaks to a problem I hear about constantly from the technology leaders we cover: the small operational headaches that pile up until nobody can see the whole picture.

Opal stands for operational AI layer, which turns out to be a clever way to describe what it does. If you’ve used ChatGPT or Claude, you know the personal version: upload a file, ask questions, get useful answers, all inside your own chat.

Lai’s point is that the value gets stuck there. Your teammates can’t see the context you’ve built, the company pays every time someone else uploads the same file, and nobody managing the organization can see any of it.

Opal makes that same capability organizational, and simpler. 

Documents, spreadsheets, audio, and video get processed once into a shared knowledge base everyone draws from. Agents get built from a library of just under 3,000 connectors to the software companies already use, and 196 AI models from more than 20 providers are all included. You can choose to use Claude, or DeepSeek, or any of the AI models to power your workflow. There are also guardrails that can be set up for an organizational level that can be programmed to mask or block sensitive information like customer details, revenue figures, or personal data. Set it up once at the organizational level, and it applies everywhere at once.

“All of the solutions that we see right now in the market are built by engineers for engineers,” Lai says. “But there’s another 99% of the world that aren’t engineers. What are we building for them?”

The people who run finance, operations, and sales, he argues, have been handed a chat window and left to figure out the rest.

Since founding Motiv in 2015, Lai has generated what the company says is more than $2 billion in enterprise value through launches, acquisitions, and IPOs. Along the way, his team kept toolkitting the basics every business and SaaS company needs, so each new venture didn’t lose time rebuilding them. 

Opal is that toolkit combined into one product with AI built through it, tested by running Motiv’s own operations on it first. 

The platform was incubated inside the studio the way Motiv incubates its ventures, and Lai says the plan is the same as it would be for any of them: spin it out as a standalone company and bring in its own CEO to run it.

The slot machine problem

The sharpest part of Lai’s pitch is about money.

“It’s not in Anthropic’s interest for you to use fewer tokens,” he says. “That’s where they make money. It’s like a slot machine.”

Tokens are the units AI companies bill by, roughly a word of text going in or coming out, and they’re the reason AI spending is so hard to predict. Every document an employee feeds a chatbot, and every answer it writes back, meters like a taxi.

Opal charges per seat instead, from a free tier through $39 and $99 monthly plans, and converts every frontier model’s pricing into a single currency where one credit equals one cent.

“Our model is fundamentally different,” says Lai. “We don’t have a margin on our credits. Every other platform wants you consuming as much as possible.”

Lai says the cost to use Claude or ChatGPT or any other AI model is a straight passthrough of what Opal pays, with no markup. And because Opal makes nothing on that consumption, it has every reason to help customers spend less, and Lai says his team works constantly to drive token costs down. He says Opal currently runs roughly seven times cheaper than teams doing the same work through separate chat subscriptions, and the savings come from two design choices.

The first is the shared knowledge base. When five employees each upload the same policy document into their personal chatbot, the company pays to have it read five times. Opal processes a document once and everyone with permission draws on that copy.

The second is caching. The first time an AI model reads a piece of content it gets billed in full, and every reread after that is far cheaper. Showing us his own usage, Lai points to cached reads running about a tenth of the cost. Opal structures every request to lean on those rates and handles the housekeeping, like busting the cache when a document changes so agents never work from a stale version.

Lai pulls up a consumption graph to show it. On June 28, the first day a set of documents was ingested, the platform ran through about two million tokens, all of it read fresh. Within days, the same work was running almost entirely on cached reads, with fresh input down to 12 tokens — a fraction of a cent. 

Compound that across every document and every employee and the savings are the product.

What technology leaders need to know about Opal

For the people who answer for what AI does inside a company, the question underneath every deployment is control. Opal’s answer is to divide the platform in two: a configuration layer and a work layer.

The configuration layer is where the setup happens. A technical team connects the company’s systems and sets the governance rules, and Lai estimates that work is only about 10% of what goes on in Opal. The other 90% is the work layer, where non-technical staff can build and run the actual workflows, and that part takes no engineering experience. Anyone can describe what they need and have Opal’s AI assist build the workflow for them, which is how Lai created the one he demonstrated for us.

And the controls speak to more than one office. 

Every agent, prompt, guardrail, and knowledge file carries version history, and changes go through a proposal-and-approval model borrowed from software engineering, so a team lead owns what gets published. For anyone who has ever wanted to know what an AI was actually doing, the audit log shows you, turning black-box AI into a digital paper trail anyone can follow.

For finance leaders who own the budget and never quite know where the tokens go, every model’s pricing converts to the same credit currency, and spend breaks down by model, project, team, and person. Budgets come with billing alerts and automatic shut-offs before a bill runs away with the expense account. 

For operations leaders, every time a workflow runs it leaves a full log of what each agent did, which tools it called, and what it pulled, so the work can be traced step by step.

An organization can also block any model across the company, whether that’s a frontier model it hasn’t vetted or one hosted in a jurisdiction it would rather avoid. Lai says Opal is hosted in a Canadian data centre, which he frames as part of the sovereignty case for Canadian buyers, and holds zero-data retention agreements with every model provider, meaning customer data can’t be used for training.

Lai is also candid about what is still a work in progress after launch. 

He says the platform doesn’t yet suggest which model to use for which job, so picking the right one for a quick creative task versus one for long autonomous work that runs overnight still takes someone who knows the models. 

Lai says that’s part of why Motiv sends in small teams of two or three engineers to help customers get set up. In the meantime, the company is building tutorials and documentation so more of that becomes self-serve, and working on features like automatic fallback to a second provider if one goes down. 

On the compliance side, its SOC 2 Type II, ISO 27001, and ISO 42001 certifications are all in progress.

What non-technical teams need to know

For the rest of us, the ones who have never opened a terminal on purpose, the product is the task list. A flow in Opal is a linear list of steps, and every step gets assigned to an agent or a person. A 20-step process can put a human at step three and step seven, with agents running everything in between.

For example, if a customer emails to cancel, an agent reads the message, pulls the account history, and drafts a reply with a retention offer it’s allowed to make. Before anything sends, it stops for a human to decide how hard to fight for that customer and what to say. Once they approve, agents take back over, apply the discount in the billing system, log the reason for churn, and update the forecast. A person spent 30 seconds on the judgment call. The half hour of clicking around three systems happened without them.

Any step can also get a reviewer. Sometimes that’s a person who approves the work or sends it back. Sometimes it’s a second agent whose only job is to challenge the first one’s output before a human ever sees it, catching problems earlier in the chain.

“Humans are good for building relationships, making decisions, having accountability, creativity, the things that agents fall short on,” says Lai. “Our approach is not let’s replace everybody with agents. It’s take the best of both worlds and integrate them so they can truly collaborate and get things done.”

Motiv is its own first customer. Lai’s team runs on Opal instead of working inside project management software, getting a daily brief of its top priorities, risks, and bottlenecks, assembled by agents pulling from the tools underneath. It’s been so effective that Lai says he hasn’t opened Notion or ClickUp in five or six weeks.

Running the studio on the platform is also how Lai found its edges. An agent can write its own steps when nobody knows at the outset how big a job is. He describes a website build where the team didn’t know how many pages they’d need. The agent set the strategy, mapped the site’s architecture, then created a step for each of the 80 pages and worked through them one by one, finishing an 80-page website in about three and a half hours.

Flows also improve with repetition. With an option called ‘Continuous Learning’ switched on, Opal learns from each attempt and any feedback people give, then applies it the next time. Lai’s point is that when a person first writes down how a job should go, they leave out the small things they know without thinking. Continuous Learning fills in those gaps over time, and Lai says his longest-running flows are noticeably sharper by the 10th or 15th attempt.

The obvious question with any AI doing work like this is what stops it from making things up. Lai’s answer is that every claim an agent makes has to trace back to something real (a document, a database, a calculation) and never the model’s own guess.

“It’s always sourced. Everything, everything is sourced,” says Lai.

The harder part is stopping an agent from inventing an answer when it doesn’t have one. Lai says his team mapped the specific moments where AI tends to guess, like doing math or looking up a number, and at those steps forces the agent to run a piece of code instead. Code returns the same answer every time, so there’s nothing for the model to make up. On top of that, every agent carries a standing instruction to flag when it’s making an assumption rather than state it as fact.

Lai says Opal is running 21 pilots across startups, mid-market companies, and a few enterprises. The gains show up as capacity, more customers served and more work finished with the same people.

At a radiology clinic, a 30-minute imaging appointment included 10 minutes of manual data entry, transcribing the results into the clinic’s records system after the scan. Lai says the pilot removed that step, which let one location go from seeing two patients per hour to three without adding staff or rooms. He’s careful to say it’s still a pilot, then points to what it could mean across the client’s 42 centres, plus about 300 more it partners with in Europe.

In another pilot, a bank is using Opal to scan provincial regulatory websites each day and flag new postings to its compliance team, work that would otherwise fall to someone checking sites by hand.

At a startup, the job was bigger: turn raw customer ideas into product and engineering specifications — a process that ran two to three weeks and pulled in several people. Using Opal, Lai says it now takes about half an hour. 

The bigger question Lai is chewing on is how you measure any of this. Today’s AI tools count activity, how many tokens got used, how many messages went back and forth, how many chats got started. None of that tells a business whether the work was any good or worth doing.

Because Opal runs whole jobs from start to finish, it knows something those tools don’t: what got produced, how long it took, and how many rounds of revision it went through. Lai wants to turn that record into a measure of the actual value the work created.

“We’re trying to redefine new KPIs of what agentic work looks like,” he says, describing a dashboard that might one day tell a leader they generated three million dollars in value in a day. “If you can do that, it becomes a no-brainer to use Opal.”

Nobody has figured out how to calculate that yet, Opal included. 

“It’s the next thing his team is building,” says Lai. And it connects back to the money argument: if Opal’s value to a customer is what it produces measured against what it costs to run, Lai wins by keeping that cost low.

Opal opened a startup program at launch today, where qualifying early-stage companies can apply directly for up to $7,000 in credits and discounts, subsidized for a year. The other route runs through accelerators, incubators, and VC firms, whose portfolio companies get access and whose programs Lai’s team helps build Opal tools for.

Where Opal fits among the AI tools you already have

There’s no shortage of AI products, so the fair question is what category this is. 

Lai’s answer sorts the field into three groups:

1. Chat assistants like ChatGPT and Claude are built for individuals. 
2. Vertical tools go deep on one function and don’t connect to much else.
3. SaaS platforms racing to add agent builders are bolting the new way of working onto systems designed for the old one.

“Their systems were never built around this new way of work,” he says. “And so they’re kind of Frankensteining it, and it’s not great. From a cost perspective and from a governance perspective, it doesn’t make sense.”

By Frankensteining, he means bolting agents onto software designed before any of this existed, so a company ends up paying for several tools that each run their own agents and hold their own separate knowledge, none of it connected.

Opal sits above all three, connecting to those tools and coordinating across them. 

Lai argues the companies with the best shot at building this, the big AI providers and platforms, never will. Doing it right would mean showing customers every rival’s model side by side and helping them spend less, and no one undercuts their own business on purpose. 

That’s his bet in a sentence: the AI industry’s incentives left an opening, and he built a company inside it.

Final shots

• If your people are already running personal ChatGPT and Claude accounts, you’re paying for the same work twice and can’t see any of it. Opal’s pitch is a floor under what’s already happening in your org, one place where the spending, the data, and the rules are visible.
• When the board asks what AI is costing and what it’s returning, “we’re experimenting” stops working. A platform that prices every job to the cent and logs how it got done gives you an answer, and it’s worth asking any vendor you’re evaluating whether they can show you the same.
• The real work coming is deciding which steps stay human. Opal makes you draw that line on purpose, step by step, and that’s a bigger call than any feature comparison. Leaders who make it deliberately will be in better shape than those who let it happen by default.