Context

Knowledge workers spend about 60% of their time on work about work: context-switching, status updates, repeated data entry, coordinating across systems. The friction concentrates in four places:

  • Noisy — too many notifications across too many surfaces
  • Siloed — disconnected systems prevent visibility and flow
  • Inefficient — repetitive manual entry of the same information
  • Inaccurate — limited trust in AI outputs without proper context

Why Slack?

Slack already holds the enterprise work graph: conversations, decisions, files, relationships, mentions, channels. That graph is the substrate LLMs and agents need to do real work for a company.

If AI is going to operate inside a business, Slack is where the context already lives.

Approach

Two months of cross-functional exploration: joint work sessions with Slack and Salesforce Cloud teams, customer intercepts at Dreamforce, and 20+ feedback sessions on early concept iterations.

Workshop board mapping pain points and agent opportunities across Sales, Marketing, CRO, and CMO personas
Cross-cloud work sessions
Two attendees at the Slack booth at Dreamforce reviewing a printed concept
Customer intercepts at Dreamforce
Video call from a customer feedback session walking through an early Slack agent prototype
Concept feedback sessions

What I argued for

The org was already moving. Internal and external demos all depicted the same near-future, a DM with an agent that gets your work done, and our work had to draft off that. The question I kept pressing was why Slack. If the answer was just “conversational entry point,” any chat surface would do.

Sketch of a three-pane Slack layout with account channels, thread with a 'Salesbot' agent, and a usage chart in the right rail
Multi-surface exploration: agents scattered across rails and threads.
Sketch of a 'Project Handoff Channel' with a generated canvas summarizing sales-to-implementation handoff
Sales-to-implementation handoff inside a channel.
Sketch of a proposal canvas being drafted by AI from RFP, previous proposals, and proposal outcomes
Canvas drafted from work-graph context.

The sharper argument: Slack’s value to agents is the work graph underneath it. The channels where collaboration already happens, the decisions and context already accumulated, the existing pattern of who-talks-to-whom that an agent can read as ground truth. Other surfaces don’t go away. They’re the substrate the conversation reads from.

That reframed the design problem. Instead of five parallel agent surfaces, Slackbot becomes the single front door, with channels, canvases, and dashboards as the destinations the conversation hands off to and the context it reads from. The conversation is the interface; the rest is the source of truth.

Vision

A near-future demo of Slack as a company’s home base for getting work done — conversation with Slackbot as the primary surface, with channels, canvases, dashboards, and agent task views as the places the work lands.

Vision: a Slack-first experience that brings together all your work
Slackbot as the conversational front door — channels, canvases, and agent surfaces as destinations.
Golden Demo — Introduction

Three persona stories

We grounded the vision in three end-to-end walkthroughs, one for each of the major Salesforce clouds. Each story was scripted and prototyped as a continuous demo.

Sales

Sales — end-to-end walkthrough
Agent querying top opportunities by amount
Querying top opportunities
Meeting prep reminder with discussion topics before a call
Meeting prep with topics
Quote approval workflow with agent-to-agent handoff
Quote approval handoff

Marketing

Marketing — end-to-end walkthrough
Slides agent building a presentation from canvas context
Slides agent from canvas
Campaign metrics from Tableau surfaced directly in channel
Tableau metrics in-channel
Marketing agent analyzing audience data and fetching guidance
Audience analysis

Service

Service — end-to-end walkthrough
CSAT metrics dashboard with trend detection in Slack
CSAT trends + anomaly detection
Expert finder matching the right person to a support case
Expert finder routing
Knowledge base article editing with approval workflow
Knowledge-base editing

Strategic impact

The work reshaped my own role: I now lead design for Slackbot, which inherited the ownership of this vision. The demo did something a spec couldn’t have — it gave a fragmented org something to point at and commit to. By the end, Slack as the interaction layer for Salesforce wasn’t a proposal anyone had to argue for. It was just what we were building.

Reflection

A vision demo isn’t a usability test or a product spec. It’s a vehicle for argument: something a fragmented organization can point at and say yes, that’s the shape we want.

Holding the why Slack question against an org already convinced of yes Slack took longer than I’d like to admit. The reframe was obvious in retrospect; while we were in it, it just felt like pushback.

The lesson I’ll carry forward: don’t start with the surfaces. Start with the use cases, watch where they collapse, and let the right surface fall out as a consequence.