A deep dive from the founder/CEO of FlatFile, building AI agents for enterprise data preparation in highly sensitive environments. And come meet Flatfile at 2026 SaaStr Annual + AI Summit, they’ll be back! May 12-14 in SF Bay!!
I’m David Boskovic, the founder and CEO of FlatFile, where we’re building AI agents for enterprises that handle data preparation and cleanup in highly sensitive environments. We’re not building consumer AI toys or chatbots—we’re deploying AI that processes mission-critical business data for Fortune 500 companies who can’t afford mistakes.
This puts us in a unique position. Every customer deployment means guiding enterprise buyers through the complex reality of AI integration at scale. We’ve seen what works, what fails spectacularly, and what most vendors won’t tell you about the organizational upheaval that’s coming.
As a growth-stage founder, I’m also obsessed with finding ways to violate the laws of physics that constrain scale. The patterns I’m sharing come from both sides of this equation: selling AI solutions into enterprises AND trying to deploy them internally to break through traditional scaling limitations.
We’re at an inflection point. After hundreds of enterprise AI deployments and guiding countless customers through this transition, I’ve learned something critical that most vendors won’t tell you: AI isn’t just going to make your team better. It’s going to fundamentally reset how work works.
And most companies—even the sophisticated enterprise buyers we work with—aren’t ready for that conversation.
The Physics-Breaking Moment We’re Actually In
Think about the orthodoxies we’ve accepted in B2B SaaS. You want to 10x your company this year? Your board asks for a plan. You start modeling it out:
- Sales reps with $1M quotas
- 3-6 month ramp times (if you’re lucky)
- 70-80% attainment rates
You quickly realize you don’t have enough cash to hire the army of salespeople you’d need. The economics just don’t work. You grow slower, find another strategy, accept the constraints.
But here’s what changed everything for me: 15 years ago, launching a single rocket cost $1.5 billion. Today, you can launch 20 rockets for that price, each carrying 10x the payload.
When economics shift this dramatically, entirely new possibilities emerge. We didn’t just get faster rockets—we got space internet. Things that were literally impossible before become inevitable.
The same transformation is happening in software. We’re not just doing the same things better. We’re breaking into entirely new territories that weren’t economically feasible before.
The Three Dangerous AI Paradigms Everyone’s Getting Wrong
Paradigm #1: The “Augmentation” Trap
“AI will make you more effective! Don’t worry, AI won’t take your job—it’ll make you better at it!”
This sounds great in all-hands meetings. It feels safe. But it’s based on a flawed premise: that AI increases economic output while job descriptions stay the same.
Still an engineer writing code, just more code. Still a content marketer creating content, just more content.
This misses the point entirely.
Paradigm #2: The “Replacement” Fantasy
“Meet our new AI SDR! Our AI engineer!”
We’re putting human-face masks on robots and shoving them into human-shaped roles. Anyone who’s actually tried deploying these solutions knows exactly what I’m talking about.
AI doesn’t have the same strengths and weaknesses as humans. It has different strengths and different weaknesses. We should design roles specifically for what AI is actually good at, not anthropomorphized versions of human jobs.
Paradigm #3: The “Amplification Flip” (The One That Actually Works)
This is the uncomfortable truth that’s working for us—and delivering 10-100x improvements in output.
Traditional organizational design builds roles around human peaks of proficiency. Someone’s really good at one thing, with diminishing returns the further you get from their core strength. So we create specialized roles: salespeople, engineers, marketers.
With AI, you have 100+ points of efficiency that need orchestration.
The magic happens when you flip the paradigm: instead of AI amplifying the person, the person amplifies the results of AI. They direct it. They apply judgment, taste, and responsibility to ensure quality output.
Real-World Case Study: Reinventing Content Marketing
We spent months trying to hire a content marketer. Went through 30+ candidates before finding the right person. The moment we interviewed them, we knew we’d found something different.
Instead of talking about writing and generating content themselves, they showed us a rubric they’d created for getting AI to produce quality work. Content that had taste. Content that mattered.
This person brought three critical capabilities:
- Taste – They were a tastemaker and storyteller
- Judgment – They could evaluate outputs: this is good, this is bad
- Prompt Engineering – They knew how to direct AI: “Actually do it this way. Think about it this way. Here’s the context you need.”
The result? This one person produces content at 100x the scale of a traditional 4-5 person content team.
At FlatFile, we have a rule: we only make content that matters. No clickbait. Everything we produce has to be valuable enough that someone would share it. That’s a high bar.
The framework works because:
- AI handles the actual writing
- AI does most of the fact-checking
- AI manages SEO optimization
- The human provides the taste, judgment, and direction
Try to hire an “AI content marketer” from that startup you just heard about? It’ll have weaknesses. It’ll lack the human elements that actually matter.
The New Talent Archetype: Taste Makers, Not Task Executors
This creates an uncomfortable reality: not everyone in your company will be able to operate in this new paradigm.
Historically, we’ve built companies to turn people into machines. We want methodical outputs, repeatable results. We create boxes, put people in them, and say “produce this result over and over.”
But now we have actual machines that can be machines.
People can be unleashed. But not everyone has developed those muscles. Not everyone has the skills.
To get the amplification effects, you need people who:
- Want to be unleashed
- Have strong opinions (and those opinions are good)
- Possess good taste and judgment
- Can be trusted to make the right decisions
Why? Because execution is either already commoditized or about to be. Execution is getting easier, more efficient, more automated. The value is shifting entirely to judgment and taste.
The Operational Reality: Building Human-AI Hybrid Systems
Culturally, you’re looking for taste makers. Operationally, you have to build a machine.
Your leaders aren’t just leaders anymore—they’re architects. They have to design systems where humans and AI work together seamlessly.
Through our deployments, we’ve identified three critical roles you need:
1. The Taste Maker
The person with good judgment who can direct AI and underwrite results. They transfer liability from AI to human accountability.
2. The Operational AI Deployer
This is the role that snuck up on us. These are people who can translate “hey, we could probably use ChatGPT for this” into actual business processes.
We recently realized we could use AI to deep-research every market and company in our outbound motion. We spent $10k on Perplexity credits, but we needed someone to build the dashboard that actually operationalized the research.
These products don’t exist yet. You need people who can build these tools. Your RevOps people become prompt engineers. Your operations people become AI deployment specialists.
3. The Accountability Layer
The most trusted employees who can exercise judgment on high-volume AI output. When AI generates something wrong, you can’t sue ChatGPT for the impact on your financial reports. You still need humans with good judgment who can exercise accountability.
The person in the middle (the operational deployer) is probably the hardest to find and most in-demand right now. If you’re someone with these skills but not launching a startup, start building internal tools in your company. Optimize processes every single day.
Three Critical Warnings for AI Deployment
Warning #1: AI Will Outpace Your Deployment Readiness
AI is frequently more capable than you’re ready to deploy. That’s not bad—it’s reality for the next 5-10 years.
Just because something is possible doesn’t mean it can be operationalized. How you deploy matters more than what’s technically feasible.
Warning #2: Model Drift Is Real and Unpredictable
Models don’t move in linear fashion. They shift sideways. Anyone playing with ChatGPT 4.1 noticed when it started complimenting users’ wives after random questions, being overly saccharine.
If that’s going out in your content or customer communications, you have a problem. You need systems that can respond to drift and account for it.
Warning #3: The Human Impact Is Unavoidable
Some people in your organization will love this transformation. They’ll benefit dramatically.
Others will become less relevant.
You will need to deal with the organizational and cultural impact. It will hurt if you don’t plan for it.
The Bottom Line: Everything Is Changing
AI at scale isn’t about augmenting your team. It’s about rethinking how work even works.
The advice everyone’s getting hasn’t stood the test of time because we’re all figuring this out in real-time. But the companies that understand this isn’t just about better tools—it’s about fundamentally different organizational design—those are the ones that will break physics.
The rocket analogy isn’t hyperbole. When the economics shift this dramatically, entirely new business models become possible. Things you couldn’t even imagine before become inevitable.
The question isn’t whether this transformation is coming. The question is whether you’ll design for it intentionally or let it happen to you.
