Form Factors for Your New AI Coworkers
A practical four-form-factor framework for integrating AI coworkers into enterprise workflows—from the team leading AI transformation at Flatfile. Discover why traditional design processes fail with LLMs and how playful experimentation leads to better AI products.
This really lets us get rid of mock-ups, get rid of the click-through prototypes and all the hand ringing about whether the thing that we're building is worth the engineering effort.
— Craig Wattrus, Flatfile • AI Engineer Summit (00:00:47)
Form Factors
Talk duration
Data onboarding AI
Real-world insights
The Four Form Factors Framework
Craig Wattrus from Flatfile presents a practical framework for categorizing AI coworkers into four distinct form factors: Invisible, Ambient, Inline, and Conversational. Drawing from Flatfile's experience building AI-powered data onboarding systems, this talk cuts through the hype to focus on what's actually working when deploying AI agents in enterprise environments.
The core insight: abandon traditional design processes (mockups, prototypes, hand-wringing) and instead "feel the material" of LLMs directly through hands-on experimentation. This approach reveals possibilities that theoretical design cannot predict.
"It's time for us to jump in and feel the material that we're working with and see what emerges."
— Craig Wattrus (00:01:03)
The Four Form Factors
Wattrus categorizes AI coworkers into four interaction patterns, each suited for different use cases and user workflows.
Invisible (Ghost)
Works silently in the background. The user doesn't need to know AI is involved—it just happens.
✓ Autonomous operation
✓ No user awareness required
✓ Results appear "magically"
"So when you sign up for a flat file, we go in the background, we take your email address, we find the company you work for, we look it up um and in the background, the AI agents are writing a flat file application."
— Craig Wattrus (00:02:18)
Ambient
Happens in the space around the user, providing subtle feedback and suggestions.
✓ Background analysis
✓ Non-intrusive suggestions
✓ User remains in control
"You can see the little sparkles pop up on the columns when it finds um opportunities to fix it. So that's ambient."
— Craig Wattrus (00:03:04)
Inline
Integrated directly into the workflow. AI operates within the user's actual work context.
✓ Direct workflow integration
✓ Real-time collaboration
✓ User maintains control
"So you're busy working in the data um and the AI is able you're able to use the AI um directly in line here um to fix the data. These agents are are writing code that runs on million-row datasets."
— Craig Wattrus (00:03:09)
Conversational
Direct dialogue/interaction with the AI—the form factor most commonly discussed.
✓ Chat-based interaction
✓ Natural language interface
✓ Complex collaboration
"And then conversational, the ones that we're um I guess all arguing about. I think that's what I learned um being here at this conference."
— Craig Wattrus (00:02:10)
Key Design Principles
Feel the Material
Move from designing through layers (mockups, prototypes) to directly experiencing LLM capabilities. Hands-on experimentation reveals possibilities that theoretical design cannot predict.
"Before with design, we were kind of looking at everything through layers - mockups and prototypes. What we need to do now is go feel the material, feel how these models work."
— Craig Wattrus (00:04:20)
Find the Grain
After understanding the material, identify where it's strong vs weak. Design interaction patterns that align with LLM strengths rather than fighting against them.
"My new north star is creating an environment for these LLMs to shine. What's this form factor that can help them nail their assignment, stay aligned, and grow as the models get better?"
— Craig Wattrus (00:04:45)
Character Coaching Over Control
Shift from controlling AI outputs through detailed specifications to designing AI personality and character through system prompts. Use tools like v0 to iterate on prompt 'character'.
"When I heard her talk, I realized I needed to go from controlling to being a character coach and actually building out the nature that I wanted."
— Craig Wattrus (00:03:30)
Design for Expression
Give AI ways to communicate its state—visual feedback, rollbacks, accountability, and even emotional signals like frustration. Enable AI to say 'I don't know' or ask for help.
"It can back off when it gets something wrong and sort of say, 'Okay, I'm handing control back over to you.' And that feels a lot better and it felt like we had found the grain."
— Craig Wattrus (00:08:50)
The "Box" Matters: Putting an Intern with a PhD in a Box
Wattrus uses a powerful metaphor: LLMs are like "an intern with a PhD." The interface you build around them is the box they live in—and the quality of that box determines their effectiveness.
"We're basically anything we do with an LLM, I feel like we're putting it in a box. You also hear people say that LLMs are like interns, like, it's an intern with a PhD. And so I think now if you're putting an intern with a PhD in a box, it better be a good box."
— Craig Wattrus (00:04:55)
❌ Putting a Formula 1 Driver in a Prius
Over-constraining AI with traditional UI patterns. Example: giving AI mouse/trackpad control felt like constraining a Formula 1 driver—it could only move one thing at a time.
✅ Finding the Grain
Designing interaction patterns that align with AI strengths. Example: multi-file context lets AI read multiple files while writing others simultaneously.
Play as Discovery: The Pelican on a Bicycle
Wattrus advocates for playful experimentation to discover emergent AI behaviors. The most valuable insights come from building "pelican on a bicycle" experiments—speculative applications that test future model capabilities.
"I think like as we find a new technology and work with it, we run the risk of just automating the tedious things. I'm most excited about those things like what emerges from playing."
— Craig Wattrus (00:09:50)
The Forward-Leaning Agent
An agent personality that is curious, excitable, and focused on getting things done. Emerged from play, not from requirements.
Unexpected Problem-Solving
Agent suggested contacting HR to get missing employee IDs—a problem it couldn't solve directly, but knew how to help the human solve.
"I was fully expecting better suggestions. I was fully expecting more suggestions. But then here the agent decided I can't fix this but I know how to fix it and so I'm going to tell you how to fix it."
— Craig Wattrus (00:13:40)
Key Takeaways
Abandon Traditional Design Processes
Mockups and prototypes are obsolete
- •Design directly with LLMs to understand possibilities
- •Feel the material through hands-on experimentation
- •Let emergent behaviors guide your design decisions
- •Don't waste time on artifacts that AI makes obsolete
Design AI Personality, Not Just Outputs
Character coaching over control
- •Use system prompts to shape AI character and behavior
- •Iterate on personality with tools like v0
- •Enable AI expression—visual feedback, rollbacks, uncertainty
- •Design for AI strengths, not human constraints
Match Form Factor to Context
Not every AI needs to be conversational
- •Invisible: Background tasks where user awareness doesn't matter
- •Ambient: Subtle guidance and suggestions in the workspace
- •Inline: Direct workflow integration for active collaboration
- •Conversational: Complex tasks requiring dialogue and clarification
Embrace Playful Experimentation
The best discoveries come from play
- •Build 'pelican on a bicycle' experiments to test limits
- •Let AI surprise you—emergent behaviors are valuable
- •Don't just automate tedious tasks—explore new possibilities
- •Use play as your primary discovery method
Design for AI Expression
Give AI ways to communicate its state
- •Visual feedback shows what AI is doing
- •Rollbacks and accountability build trust
- •Enable AI to say 'I don't know' or ask for help
- •Emotional signals (frustration, uncertainty) improve UX
Prepare for Constant Change
Designs will need frequent rebuilding
- •Models improve rapidly—designs must evolve
- •Build flexibility into your form factors
- •Stay humble—your perfect design will need changes
- •Learn from each iteration and don't get attached
Key Moments
Form factors for your new AI coworkers
Craig Wattrus • Head of AI Transformation, Flatfile • AI Engineer Summit
Research Methodology
All quotes in this report are timestamped and link to exact moments in the video for validation. This analysis covers Flatfile's four-form-factor framework for AI coworkers, material-first design philosophy, character coaching approach, and the role of playful experimentation in discovering emergent AI behaviors.