Agentic Workflow - Kiran Tanwar Portfolio 2026

Case Study · GTM Buddy

Agentic
Workflow Builder

Build Workflows That Think

An AI-powered workflow builder that helps Sales, CS, and Enablement teams create autonomous workflows using agents and tools that understand context, reason through next steps, and execute tasks automatically.

Role: Product Designer

Type: Enterprise SaaS, AI-first, Workflow Automation

Team: Product, Data Science, Engineering, Frontend

Agentic Design

AI-First UX

Enterprise B2B

UX Research

Manager

Trigger

Event

Agent

Research

Tool

Fetch Data

ARE

Reasoning

Execute

Auto-Run

Agentic Workflow Canvas

Overview

What is Agentic Workflow?

Agentic Workflow is an AI-powered system that lets teams build workflows visually or through intent-based input. Users drag and drop agents and tools, while ARE - Autonomous Reasoning & Execution - helps the system understand context, suggest next steps, and execute workflows intelligently.

Build

Drag and drop agents and tools onto a visual canvas to compose workflows without engineering dependency.

Understand

AI reads the workflow context to understand intent, data dependencies, and what comes next.

Reason

ARE suggests next steps, resolves ambiguity, and helps users make smarter decisions as they build.

Execute

Workflows run automatically once published - no manual triggering or task handoffs required.

Improve

The system learns from outcomes, surfaces optimization suggestions, and adapts to changing business context.

The Problem

Automation Was Still Too Manual

Sales and enablement teams were spending hours creating manual workflows across scattered tools. Existing automation was rigid, hard to scale, and broke the moment processes needed reasoning, adaptation, or context.

70%

Workflows Fail

Due to rigid rule-based logic that can't adapt to changing inputs or business context

Industry research

3–5 hrs

Manual Setup Per Flow

Time spent wiring each workflow from scratch across disconnected tools and systems

User interviews · n=15

guidance

No In-Product Help

Users had no access to suggestions, templates, or AI assistance while building workflows

Heuristic evaluation

Why This Matters

Broken Workflows Create Operational Drag

When workflows depend on manual setup, teams lose time, automation becomes fragile, and users need technical help for small changes. For Sales, CS, and Enablement teams, this slows down execution and makes scaling harder.

For Sales Ops

Too much time spent syncing CRM fields, preparing summaries, and fixing broken automations that should run themselves.

For Enablement

Difficult to scale onboarding, content prep, coaching notes, and training workflows without engineering support.

For Customer Success

Follow-ups, onboarding steps, and account health checks remain scattered across tools with no unified flow.

For the Business

Teams move slower because workflows automate actions, but do not reason or adapt when conditions change.

Goals

What Success Looks Like

Visual + Intent-Based Creation

Let users build workflows through drag-and-drop or by describing their goal in plain language.

Reusable Agents and Tools

Create intelligent building blocks that teams can share and reuse across Sales, CS, and Enablement workflows.

ARE as the Core Layer

Introduce Autonomous Reasoning & Execution to help workflows think before they act - not just automate steps.

Adaptive Execution

Move beyond static automation into workflows that adjust based on context, data, and outcomes over time.

Research

Understanding How Teams Build Workflows Today

I started with product discovery to understand how users currently create workflows, where they get stuck, and what kind of AI assistance would actually help without removing control.

Workflow Builder Audit

User Interviews

Shadow Sessions

Heuristic Evaluation

Workflow Builder Audit

Benchmarked Zapier, n8n, Relevance AI, and Airplane.dev to map capability gaps, interaction patterns, and what users praised or complained about in each tool.

User Interviews

Spoke with 15+ users across Product, Sales Ops, and Enablement to surface workflow pain points, mental models around automation, and unmet needs that existing tools couldn't address.

Shadow Sessions

Observed users building workflows manually in HubSpot, Gong, and ChatGPT. This revealed the workarounds, copy-paste habits, and invisible cognitive load that interviews alone couldn't capture.

Heuristic Evaluation

Tested existing GTM Buddy modules against Nielsen's heuristics to identify where cognitive overload occurred, where users lost context, and what conventions needed to carry forward.

What Users Told Me

I get lost trying to connect tools. What if the system could suggest what comes next?

It takes too long to build a workflow from scratch every time.

I want to describe my goal, and the system should figure out the rest.

Competitive Audit

What Existed vs. What Was Needed

Most workflow tools supported automation, but were either too generic, too technical, or lacked sales enablement context. The opportunity was to create a hybrid builder that combined visual control, natural language input, and AI reasoning.

Capability	Zapier	n8n	Relevance AI	GTM Buddy Agentic Workflow ✦
Workflow Creation	Strong trigger-action automation, but mostly rule-based.	Flexible visual builder, but more technical for non-technical teams.	AI-first workflow setup, but less focused on enterprise sales workflows.	✓ Combines visual canvas with intent-based workflow creation.
Context Understanding	Executes predefined steps, but does not understand sales context.	Supports logic and conditions, but context must be manually configured.	Can use AI agents, but context depends heavily on setup quality.	✓ ARE reads workflow context, user intent, data, and business logic before acting.
Agent/Tool Selection	Large app ecosystem, but users must know which tool to choose.	Powerful nodes, but discovery can feel overwhelming for new users.	Agent setup is flexible, but requires understanding of AI components.	✓ Categorized agents and tools guide users toward the right building blocks.
Reasoning and Execution	Automates actions, but does not reason through next-best steps.	Can support branching, but logic is entirely manually wired.	Supports AI agents, but reasoning visibility and control are limited.	✓ ARE reasons, suggests next steps, executes, and adapts the flow.
Sales Enablement Fit	Generic automation across many industries.	Generic workflow automation built for technical users.	AI workflow platform, but not deeply sales-enablement specific.	✓ Designed specifically for Sales, CS, and Enablement workflows.
Visibility and Control	Users can view task history, but not AI reasoning.	Good execution logs for technical users only.	Agent outputs are visible, but reasoning may not be clear.	✓ Shows workflow status, run history, failure points, and ARE reasoning logs.

Existing tools solved parts of the workflow problem - automation, visual building, or AI agents. But most were either too generic, too technical, or too dependent on manual setup. GTM Buddy's opportunity was to create a sales-enablement-first workflow builder that combines visual control, natural language intent, contextual recommendations, and autonomous reasoning through ARE.

Personas

Three Roles, One Workflow System

Agentic Workflows needed to support the people who create, manage, and benefit from automated sales enablement operations. Each role shaped how the system handled speed, control, visibility, and trust.

Sales Enablement Manager

Core Job

Creates onboarding, coaching, content prep, and learning workflows without depending on Engineering.

Key Need

Build without engineering, automate content prep, see what's running, and scale without adding headcount.

Frustration

Repetitive tasks, rigid workflows, and over-reliance on technical teams for small changes.

Sales Operations Analyst

Core Job

Builds CRM, reporting, meeting prep, and data workflows that need accuracy, visibility, and repeatability.

Key Need

Multi-system automation, agentic reasoning, clear execution logs, and the ability to preview before releasing.

Frustration

Tedious CRM syncing, complex multi-tool workflows, and no way to diagnose failures quickly.

Customer Success Leader

Core Job

Uses adaptive workflows for onboarding, renewals, QBR prep, follow-ups, and account health monitoring.

Key Need

Automated onboarding sequences, context-aware follow-ups, and agentic flows for renewal management.

Frustration

Onboarding scattered across systems, repetitive manual tasks, and no unified view of what's running.

These roles helped define how the builder should suggest steps, automate repetitive tasks, maintain consistency, and support fast execution across teams.

User Journey

From Idea to Running Workflow

Open workflow listing

Create new or use template

Choose trigger

Add agents and tools

Configure each step

Test and preview

Fix errors

Save and publish

Monitor runs and logs

The journey was designed to reduce blank-canvas anxiety. Users could start from a template, build visually, test safely, and understand what happened after the workflow ran - without needing a technical background.

Information Architecture

Structuring an Agentic Workflow System

The IA was structured around how users move from intent to execution: define the workflow goal, choose triggers, add agents and tools, configure logic, test safely, and monitor what happened after launch.

Agentic Workflow Module

Workflow Listing

Search & Filter

Find workflows fast

Favorites

Quick access

Status + Last Run

Health visibility

Owner + Run Count

Accountability layer

Quick Actions

Run · Clone · Archive

Workflow Canvas

Trigger Node

Entry point

Agent Node

Reasoning block

Tool Node

Data / action

Condition Node

Logic branching

Configure Drawer

Node settings

Auto-save

Draft protection

Core Intelligence Layer

ARE - Autonomous Reasoning & Execution

The most important IA decision was to make ARE visible as a system layer, not a hidden backend feature. Users needed to understand what the AI suggested, why it acted, and where they could still control the workflow.

Reads context Understands dependencies Suggests next steps Resolves missing inputs Executes the flow Explains reasoning Adapts on outcomes

Testing + Execution

Preview Run

Safe test mode

Publish

Go live

Pause / Resume

Runtime control

Rerun

Retry on failure

Logs + Insights

Execution History

Run Timestamps

Status per Step

Failure Points

ARE Reasoning Logs + Performance

Reasoning Trail

Why path was taken

Context used

Confidence signal

Alternative paths

Workflow Performance

Completion rate

Avg run time

Error frequency

Step drop-off

Optimization Signals

Suggested improvements

Bottleneck detection

Entry Point

System Layer

Module

Feature

Data flow

ARE Layer

Design Decisions

Solving the Hard Problems

Workflows as Operational Assets

Instead of hiding workflows inside settings, I designed a clear listing page with status, runs, owner, last run, and favorites. This made workflows feel manageable and trackable - not invisible backend configurations that teams lost track of after setting them up.

Agentic Workflows listing page showing favorites, run counts, and status

Templates Before Blank Canvas

Users struggled to start from scratch, so I added templates and favorite workflows to reduce setup fatigue. The goal was to help users start faster without losing flexibility - a blank canvas is not empowering, it's paralyzing. Templates gave users a launchpad, not a constraint.

Browse Templates gallery showing pre-built workflow templates by category

Visual Builder + Intent-Based Input

I combined drag-and-drop creation with AI-assisted suggestions. Users could either build manually or describe their goal and let the system suggest the workflow structure. Power users kept control; new users got guidance. The same surface served both without switching modes.

Workflow canvas with drag and drop node placement

Clear Node Categories

Triggers, agents, tools, and actions were separated in the left rail with distinct icons and color coding. This reduced the "where do I find X?" problem and helped users understand what each building block does before dropping it on the canvas.

Workflow canvas showing categorized sidebar with Triggers, Agents, Tools, and Actions

Configuration Without Losing Context

I used side drawers for trigger and node configuration so users could edit details while still seeing the full canvas. This kept the mental model intact - users always knew where they were in the workflow while making changes to individual steps.

Configure Trigger side drawer open while workflow canvas remains visible

Pre-Built Agents to Start Faster

Instead of building every agent from scratch, I gave users a library of pre-built agents they could add straight into a workflow. This lowered the barrier to a first working automation and let teams assemble capable workflows from proven building blocks rather than a blank canvas.

Add pre-built agents panel showing a library of ready-made agents to drop into a workflow

Show the Reasoning Trail

ARE needed visibility, not mystery. I designed logs and reasoning views so users could understand why a workflow ran, why it failed, or why it chose a specific path. This was not optional - trust in an agentic system depends on explainability at every step.

Logs and Insights dashboard showing run history, ARE scores, optimization signals, and failure points

Other Design

Exploring the Full Workflow Experience

Beyond the core canvas, I explored supporting screens that helped users discover workflows, configure agents, test execution, and understand what happened after the workflow ran.

Agentic Workflows empty state with centered CTA to create a new workflow or start from templates

Empty State - clear onboarding prompt with two paths: start from scratch or browse templates

Agentic Workflows listing page showing all workflows with status, run count, last run, and owner

Workflow Listing - a central place to view all workflows with status, owner, last run, and quick actions

Agentic Workflows listing page with pinned favorites row and full workflow table below

Workflow Listing with Favorites - pinned workflows surfaced at the top for fast access alongside the full table

Workflow canvas with status bar showing Inactive toggle, Test/Preview, Run, and Publish

Workflow Canvas - active state with Inactive/Active toggle, Test/Preview, Run, and Publish controls visible in the top bar

Workflow canvas drag-and-drop state showing ghost node placement

Workflow Canvas - a visual builder where users connect triggers, agents, tools, actions, and outcomes

Configure Trigger side panel open alongside the workflow canvas

Configure Trigger Drawer - a focused side panel to set trigger details without leaving the canvas

AI Agents listing page showing all agents with category, run count, status, and quick actions

AI Agents - a categorized library of reusable agents with status, run history, and team ownership

Content Copilot agent detail page showing workflow canvas with performance metrics panel

Agent Detail View - a dedicated canvas per agent showing connected nodes, run performance metrics, and the full node library in the left rail

Agent workflow canvas with node sidebar showing Core and Content Generation node categories

Node Library - categorized left panel with Core nodes (Auto-Tag, Coaching Notes, Knowledge Check) and Content Generation nodes for building agent workflows

Logs and Insights - run history, errors, ARE reasoning logs, and suggestions to improve the workflow

These supporting designs helped turn the workflow builder from a canvas-only experience into a complete operating system for creating, testing, and improving agentic workflows.

Impact

Early Testing Signals

40+

Feedback Notes

Collected and incorporated into active iterations

Successful Runs

Workflow completions during early testing

62%

Active Testers

From Sales Enablement, CS, and Sales Ops actively tested and gave feedback

3–5 min

Draft Build Time

Average time to build a draft workflow in early testing

The project is still in active iteration, but early testing showed strong interest in guided workflow creation, reusable agents, and visibility into execution - confirming that ARE visibility was the right design bet.

Learnings

What I Learned Designing Agentic Systems

AI needs visibility.

Users do not trust automation unless they can understand what happened, why it happened, and what to fix. Visibility is not a nice-to-have in agentic systems - it is the product.

A blank canvas is intimidating.

Templates and AI suggestions are not shortcuts - they reduce decision fatigue and help users build with confidence. The best starting point is always a structured example, not an empty space.

Control matters in agentic UX.

When AI can act automatically, users need preview, pause, edit, and rollback options before they feel safe using it. Autonomy without control creates anxiety, not efficiency.

Reasoning is a UX layer.

ARE is not only a backend capability - it needs to show up in the interface through suggestions, logs, explanations, and confidence signals. If the reasoning is invisible, the product feels like a black box.