Project Overview

As AI agents gain the ability to act, people need a clear way to direct them. Relay explored that interface with a shared workspace for understanding what multiple agents are doing, shaping their work and knowing when to step in. The goal was to support AI researchers, product managers and sellers coordinate many agents across disconnected systems, with clear progress, traceable sources and simple points of control.

The product exploration points toward an emerging modality of how agents swarms as big as 100+ are managed, steered and explainable when achieving ambitious goals and briefs.

Owned:
→ Problem framing and hypothesis definition
→ Research design and synthesis
→ Supervision architecture for agent tasks
→ Front-end engineering
→ End-to-end prototyping
→ Executive alignment and roadmap decisions
→ Evaluation framework for agent reliability and supervision quality

Context

AI was moving beyond conversation into work that spans tools, data and time. But enterprise context remained scattered across systems, with inconsistent data and constantly changing state. Relay explored one surface where researchers, product managers and sellers could bring that work together.

Delegation only helps when it reduces the effort of supervision. People need to see progress, reconcile conflicting sources and steer work as it unfolds, without becoming full-time managers of automation.

The Problem

The project centred on one question:

How can people delegate long-running work to multiple agents while keeping a clear view of progress, decisions and control?

Early conversations with technical and non-technical users revealed a recurring cost: coordinating work across tools. Switching systems, stitching context together and carrying it into the next task consumed the time agents were meant to save.

The design constraints were sharp:

→ Agents could only help if users could understand what they were doing and why.
→ Any solution that increased setup complexity would fail adoption.
→ Without clear intervention and recovery, autonomy creates risk and hesitation instead of leverage.

Our Persona(s) Flows for Relay

Design Opportunity

The design opportunity was to make delegated work inspectable. A task needed a visible purpose, a record of actions and a clear place for a person to intervene.

I framed the design space as:

→ Agents could only help if users could understand what they were doing and why. Turning invisible background automation into something observable and accountable.
→ Giving users control without forcing them into micromanagement. Making uncertainty explicit (freshness, source reliability, execution confidence).
→ Designing for recovery as a default, not an edge case.

Exploration

I started by mapping end-to-end workflows and identifying where coordination broke down across systems. From research synthesis, I formalised a model to structure agent work:

Tasks → Intents → Actions

→ Task: the user-recognisable unit of work
→ Intent: what success means and what constraints apply
→ Actions: concrete steps executed across tools and data sources

Concept Exploration: Supervision Models.

This structure wasn’t only conceptual, it became a design tool for defining what should be automated, what should be confirmed, and where escalation is required.

To stress-test the model, I ran adversarial prompt testing using ambiguous and conflicting requests, stale context, and scenarios that should trigger escalation rather than execution. These tests helped define escalation rules: when the system should ask, confirm, or act and what evidence it must surface before doing so.

In parallel, I explored three high-level supervision models:

Modular: a workspace composed of widgets representing context, tasks, and monitors
‍Unified: a single end-to-end workflow experience for completing work in one place
‍Transformative: cross-system analysis and synthesis as a primary value driver

User testing and stakeholder review pushed the direction toward a system that was modular in structure, but guided in configuration, reducing blank-canvas burden while retaining flexibility.

Based on our research findings, we evolved our modular dashboard concept but three new design challenges emerged. These needed to be reshaped after testing and evaluation.

Project Relay, 2026

What the research changed

1. “Spaces” for Separate Accounts

Hypothesis: sellers benefit from switchable account surfaces.
‍Finding: context-switching increased cognitive load and felt like managing browser tabs.
‍Decision: move to a sidebar holding ‘projects’ or ‘space’ to manage specific content.

Failed spaces tab - Relay.

2. Parallel Agent Tabs

Hypothesis: running several agent tasks in parallel tabs would let users supervise more work at once.
Finding: with state hidden inside each tab, users lost track of which tasks were running, blocked or done.
Decision: consolidate parallel work into a single Tasks view, with explicit state transitions and one place to supervise progress.

3. Manual Integration Onboarding (Packs &/skills)

Hypothesis: Configuring systems, widgets and platform integrations is challenging.
‍Finding: Setup friction stalled activation, especially for non-technical users.
‍Decision: introduce pre-configured Packs & Skills tailored by role. Auto-generate widgets on behalf of users. This was a significant change in the structure to the supervision model.

Design learnings

1) Autonomy without observability increases cognitive load

Even when output quality was high, users hesitated if they couldn’t see progress, dependencies, or what the agent was basing decisions on.

2)  Invisible work is risky and reduces trust

Plausible results built from stale context were more trust-damaging than obvious errors, because users couldn’t detect them early. Freshness and provenance must be designed, not assumed.

3) Long-running tasks require a legible state model and check-ins

Users struggled to supervise parallel threads when state was implicit. Consolidating into a single Tasks view with clear transitions reduced confusion and improved confidence.

Future Directions

This exploration points toward an emerging idea of how agents swarms as big as 100 are managed and steered successfully to achieve ambitious goals. As models become more capable, product differentiation shifts to supervision quality interfaces that make agent behavior:

→ Observable: users can see what’s happening and why
→ Interruptible: users can intervene without breaking the system
→ Accountable: actions have traceable evidence and history
→ Recoverable: the system supports safe rollback and clear next steps

The next challenge is an interaction model that scales with the number of agents. People need to understand what changed, what remains uncertain and which decisions are theirs to make.