Building an Enterprise AI Productivity Platform: From Slack Bot to Integrated AI Workforce

toqan · Tech · 2025 · source Agentic: yes · promotional bias: med · source-backed · extracted with prompt 0.5.4 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Enterprise AI productivity adoption

Toqan needed to evolve from a Slack bot into an enterprise AI productivity platform serving a large portfolio organization with many companies, employees, models, integrations, APIs, and bottom-up adoption needs.

Constraints:

The organization spans roughly 100 portfolio companies and about 30,000 employees.
Users need both conversational Slack access and API/system integrations.
Adoption depends on training, feedback loops, and super-user behavior, not only model access.
Enterprise workflows need authentication and API-layer controls.

Architecture

Toqan started as a Slack application giving employees access to multiple LLMs, then expanded toward an integrated AI workforce with APIs, workflows, authentication, and bottom-up agent creation. The rollout combined hackathons, workshops, training, Slack community feedback, and metrics such as daily active users, super users, API calls, and workflow adoption.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  employee[Employee] --> slack-app[Slack App]
  employee --> api-layer[API Layer]
  slack-app --> toqan-platform[Toqan Platform]
  api-layer --> toqan-platform[Toqan Platform]
  toqan-platform --> model-access[Model Access]
  api-layer --> enterprise-systems[Enterprise Systems]
  toqan-platform --> adoption-analytics[Adoption Analytics]
  feedback-community[Feedback Community] --> adoption-analytics[Adoption Analytics]

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`Employee`	Enterprise user asking questions or running workflows.	`Slack App`, `API Layer`
`Slack App`	Conversational interface for model access and collaboration.	`Employee`, `Toqan Platform`
`API Layer`	Workflow and system integration surface.	`Employee`, `Toqan Platform`, `Enterprise Systems`
`Toqan Platform`	Enterprise AI productivity layer combining models, workflows, and adoption telemetry.	`Slack App`, `API Layer`, `Model Access`, `Adoption Analytics`
`Model Access`	Access to multiple LLMs and future agent capabilities.	`Toqan Platform`
`Enterprise Systems`	Portfolio company tools and endpoints exposed through APIs.	`API Layer`
`Adoption Analytics`	Measures daily active users, super users, API calls, and workflow adoption.	`Toqan Platform`, `Feedback Community`
`Feedback Community`	Slack channel, workshops, hackathons, and trainings that shape adoption.	`Adoption Analytics`

Data flow

Employees interact through Slack or APIs. Toqan routes requests to model and workflow capabilities, layers user-level authentication over APIs, tracks usage and adoption metrics, and uses a large Slack feedback channel plus trainings to discover and improve workflows.

“models available through Slack for people to kind of experiment with them and eventually that grew out to be token and that's now today it's a uh basically a productivity tool and even starting to”transcript, stated

“our assumption is that the uh user level authentication on top of a layer of APIs is good enough for many workflows.”transcript, stated

Why this architecture

Toqan treats enterprise AI adoption as a socio-technical rollout: the platform offers Slack and API surfaces, but training and community feedback make people use it. Bottom-up creation is emphasized because practical workflows emerge from users closest to the work.

“we did lots of hackathons and workshops and training sessions. Um, we have the benefit that in reality we have a pretty contained small audience, right? I think we we look at about 30,000ish employees.”transcript, stated

“agent creation needs to be bottom up for the most part. Um it can be quite difficult to beyond examples share hey this is relevant for you because um someone might work in Tableau you might”transcript, stated

Models used: multiple LLMs

Scale

The broader group includes about 100 companies.
“about 100 companies part of it. Some very large like ox and swiggy and i food and delivery hero and others and some smaller. But you know at the center here at the AI team uh that we have built”transcript, stated
The rollout considered about 30,000 employees.
“we we look at about 30,000ish employees. And so I'm not >> not so small. >> No, no, but I'm not too worried of like, oh, we will have a potential of I don't”transcript, stated
The feedback Slack channel had about 1,300 users.
“Slack channel that uh has 1,300 users in there that if we have an idea, we have a question, we just chuck it in there and we get lots of feedback right away. Um but what we've learned is build a first version as quickly as possible. Like the”transcript, stated
Some platform APIs expose 100-plus endpoints, with a subset selected first.
“we're looking at 100 plus endpoints that the platform can have. We've selected what we think is the most relevant for now”transcript, stated

Alternatives considered (roads not taken)

Slack-only assistant.

Rejected because: Users also needed workflow/API integration beyond conversation.

“token not just in Slack but also you know in a web interface if people want to chat to it and so on and now also through APIs”transcript, stated

Top-down use-case rollout only.

Rejected because: The team learned agent creation should be bottom-up.

“AI employees are going to be bottom up and that inherently means bottom up and uh very very narrow in scope and again micro but and um because of the AI employees that we've been looking at as”transcript, stated

Expose every possible endpoint immediately.

Rejected because: The platform selected the most relevant subset first to manage overload.

“example. Um an an overload of endpoints, right? So we're looking at 100 plus endpoints that the platform can have. We've selected what we think is the most relevant for now um and build tools”transcript, stated

Patterns

`tool-use-orchestration` (enterprise-workflows)

Toqan exposes workflows through APIs and Slack rather than only chat.

“workflows through an API and there we measured API calls”transcript, stated

`observability-tracing` (adoption)

The platform measures daily active users, super users, API calls, and workflow adoption.

“super users who is that's basically the”transcript, stated

`human-in-the-loop-gate` (adoption)

Hands-on trainings and user feedback loops guide what workflows become production capabilities.

“we speak to users a lot”transcript, stated

Tradeoffs

Adoption strategy

Chose Bottom-up usage and agent creation over Only centrally designed workflows.

Rationale: Practical workflows emerge from users and portfolio companies.

“agent creation needs to be bottom up”transcript, stated

Platform surface

Chose Slack plus API integrations over Slack bot only.

Rationale: Real productivity requires integration into systems, not just conversation.

“through APIs”transcript, stated

Gotchas (what bit them)

Enterprise AI adoption is not solved by access alone.

Root cause: Users need training, workflow fit, and community feedback.

Resolution: Run hackathons, workshops, training sessions, and feedback channels.

“training and guidance and materials.”transcript, stated

Too many integration endpoints can overwhelm early workflows.

Root cause: Enterprise systems may expose 100-plus endpoints.

Resolution: Start with the most relevant selected tools and expand from feedback.

“We've selected what we think is the most relevant for now”transcript, stated

Evaluation

Toqan evaluates through adoption metrics such as daily active users, super users asking five-plus questions per day, API calls, workflow usage, and feedback from a large Slack community.

Metrics: Daily active users · Super users · API calls · Workflow adoption · Slack feedback · Training participation

“productivity and adoption of Genai”transcript, stated

Outcomes

Toqan expanded from Slack bot toward enterprise AI workforce platform.
The rollout targeted about 30,000 employees across around 100 companies.
A 1,300-user Slack channel supplied rapid product feedback.

The source cites about 100 companies, roughly 30,000 employees, 1,300 Slack-channel users, 100-plus platform endpoints, and super-user/adoption metrics.

“build a first version as quickly as possible.”transcript, stated