One morning, a couple weeks back, a query from a client had been sitting inside a Figma thread for close to 19 hours. It concerned a screen our team was actively designing in the middle of a live sprint. Several replies had gathered around it, but the conversation still had no decision. I tend to keep a close eye on Figma comments; even then, an active studio can produce enough of them to make notification badges fairly useless.

What I did next was fairly straightforward:

• Opened Figma and answered the question.

• Spoke with the design manager and designer.

• Closed the loop before our day had even properly begun.

Beginning the day at a design studio where the team works geographically distributed, across cities, comes with its set of challenges. Figma has comments across active files. Slack and email carry the conversations around these files, the documentation is maintained across Notion, while reviews and project updates continue to move. Before we get into the day’s work, somebody has to work out what needs attention first.

Every morning, a briefing assembled from the tools we use lands in my inbox. It groups what changed while I was away, by project, and flags the items that need my attention. That morning, the unanswered Figma thread appeared near the top. It had been marked as urgent, with enough of the surrounding exchange to show why it needed me.

In the morning briefing that lands in Premankan's inbox, the 'Needs your attention' section shows items ordered by urgency, each labelled by the action required and routed back to its Gmail or Slack thread.

This briefing is one of the more visible outputs of a system I have been building around my work. The system follows what is happening across our projects and retains the surrounding context, allowing an unresolved thread to return before I have thought to ask about it. I think of this quality as ambient intelligence because most of the useful activity happens in the background, around the work we are already doing.

Further into the same morning briefing, quieter activity across Business development and Studio / content stays tracked in the background, from proposals and contracts to articles and deliveries.

To return to the Figma thread, a regular notification would have shown me that the comment existed, where it could easily have joined several others. The morning briefing could place it inside the larger project conversation, recognize that the problem remained unanswered, and put it ahead of everything that could wait.

That small intervention depends on a much wider view of the work. Here is how we are building that view into the system, and what we need to resolve as we extend it across Canvs.

1. Reading the project as a whole

Connected tools (either via MCPs or APIs) provide a continuous record of project activity. To augment that, I have a local intelligence layer for the parts that happen between their updates, including what is said during a call and what is visible on my screen at the time.

On my machines, this layer transcribes system audio and recognizes & indexes text that appears on screen. If I am in Figma while speaking with a designer, it can retain both the conversation and the context of the interface we were discussing.

The whole system in one view, with signals becoming shared context in a common record, a person deciding what happens next, and their corrections returning to the record.

Together, these sources feed a common record. Each new item moves through the same basic process:

• Information enters through an integration with one of our project tools or through the local intelligence layer.

• The system attaches the original source, project, people involved and related file.

• The usable text and its surrounding details are added to a local knowledge base, with a route back to where they came from.

• My agents can draw from that record for their own work and return what they produce to the same place.

What I have is a fleet of agents working around the same context. While one conducts research in the background, another cross-references the output of that research against existing context in my system, while a third agent prepares a briefing based on the synthesis of the first two.

When a research job finishes, its result becomes part of the record. The agent assembling my next briefing knows that the work happened and can surface it when it becomes relevant to a project.

With earlier work and current activity available in the same layer, the system can recognize when a decision from one moment belongs to another.

In one case of new research meeting existing context, BNPL India was flagged as a gap a month earlier, then investigated by an agent whose findings updated four briefings and reframed the studio's posture.

2. Knowing when an earlier decision matters again

A recent example began during a design review. While reviewing some designs in a Figma file with one of our designers, we discussed a constraint in the project’s visual language and arrived at a decision.

Several days later, one of our design managers encountered the same constraint in a separate client thread. When the new Figma comments entered the common record, the system connected them with my earlier call and brought that discussion into my morning briefing.

The original transcript remained attached to the meeting note, so we could check the connection and recover the reasoning behind the decision. I could share the source with the design manager or jump in to the client thread with the earlier discussion already in mind.

A design review produces more than the direction we eventually choose. We explore other routes and set them aside for specific reasons. Those reasons can surface again when a similar suggestion returns several weeks later.

By the time this constraint came back, we had already done part of the thinking it required. The system made that thinking available to the people handling it now, so the conversation could continue from the point the project had already reached.

3. Deciding what deserves to interrupt

An ambient system makes a judgement each time it moves something from the background into the active context. Like any system that evolves over time, those judgements were not always right.

A team member once sent me a Slack message about taking a few days off. The message contained a rhetorical question for humorous effect, and I responded with a checkmark reaction on Slack, as I tend to do all day long on Slack and other messaging channels. The exchange was complete to the both of us – but the system saw a question without a written answer.

Two weeks later, after the person had long returned, it kept insisting that the request needed a response. When similar misunderstandings appeared, I added a rule that moved leave, holidays and other administrative updates into a separate summary.

The system now works within a few clear boundaries:

• Routine administrative information stays available without competing with active project work in the morning briefing.

• Anything it surfaces carries a route back to the transcript, message or file behind its interpretation.

• The agents can read connected sources and prepare material, but they cannot send an email, post to Slack or reply to a Figma comment on their own.

• They can evaluate a decision, proactively give me a suggestion, but they are never allowed to publicly authorize a decision on my behalf or communicate my intent as if they were me.

Workplace language gathers meanings that are particular to the people using it. A checkmark emoji can close a conversation even when nothing in the text says that it has. Each correction helps the system understand more of that surrounding practice.

The system can bring something forward and suggest what may need doing. A person still decides whether it deserves action.

Part of what makes an interruption useful is knowing when to appear and when to leave something in the background.

Information still needs curating however, and on my machines, the local layer removes raw screenshots and recordings after a set period, while retaining the text that may remain useful. This preserves project context without keeping every source artefact indefinitely. Not to mention, it saves me from having to procure ever increasing amounts of storage (and if you know the state of the storage market right now, you’ll agree this is easier said than done).

Care around what the system surfaces and retains becomes even more important as the capability reaches more people across Canvs. Each person needs access to the context their role requires while retaining control over anything the system does in their name.

4. Opening each tool with a reason

Once the surrounding context is prepared in advance, a lot of the interfaces and processes around work start to change.

One place this becomes visible is onboarding. An agent can assemble the material someone needs from the project history, giving them a way into the work that reflects their role instead of asking them to read and understand a huge ‘knowledge-transfer packet’ full of documents and notes that may or may not be relevant anymore.

The same applies to project dashboards. They continue to hold a useful record, while the system can check their status against the work in Figma and the conversations around it. I need to visit them less often simply to reconstruct where a project stands.

A meeting transcript passing through cluster, resolve, assign and trace to become a structured record, with multiple decisions and action items each linked back to their moment in the transcript.

Reminders can carry more context too. ‘Talk to Arjun (Rajkishore, Principal at Canvs)’ is not especially helpful when the reason for the conversation has disappeared. The system can bring the subject and the relevant material along with the reminder.

Figma and Slack remain important because the design and conversation happen there. What becomes less central are the additional interfaces whose main job is helping us remember where to look.

By the time I open a file or conversation, I already know what needs me and why.

Intelligence is part of running the studio

We recognize intelligence in somebody who can hold the larger picture and know which part of it matters now.

In a studio, that ability is cumulative and grows through long familiarity with projects. People remember why a direction was set aside and carry the texture of earlier conversations into the next decision.

An ambient layer gives that familiarity another way to travel. Every project adds to what we notice the next time a similar problem appears, and an ambient layer can keep more of that experience available as the work moves on.

The Figma thread we began with shows why we are building ambient intelligence at Canvs: to give the studio a form of shared intuition that draws on what it has learned at the moment it matters most.