Are You AI Fluent? Is Your Organization?

Software was supposed to make work easier. Instead, most teams are buried under it.

That’s SaaD – Software as a Disservice. Dozens of disconnected tools waste time, duplicate work, and inflate costs.

Rippling changes the story. By unifying HR, IT, and Finance on one platform, Rippling eliminates silos and manual busywork.

Companies like Cursor, Clay, and Sierra have already left outdated ways of working behind – gaining clarity, speed, and control.

Don’t get SaaD. Get Rippling.

Is SaaD hurting you?

Hello, fellow humans! As we think about how our own brains work, we can dig deeper into what we want and need AI to do, to give AI better instructions, and how to understand the outputs and act based on what AI gives us in return.

What does that mean for us as individuals? What does that mean for organizations?

Nate B. Jones produces high-quality AI news, insights, analysis, and advice at a blistering pace. For anyone doing advanced AI work, I recommend giving him a follow on YouTube and subscribing to his Substack.

He recently did a piece on the difference between being AI-trained and being AI fluent. And the difference really comes down to human judgement skills. He goes into some depth explaining what each of these mean, but here’s a quick rundown:

Check out the full video on YouTube, because he gives a lot of important context on the problem space in corporate AI environments; there are significant organizational factors and considerations that he lays out thoughtfully.

The product person in me wants to test and validate his claim that 10 AI-fluent people can outperform 500 AI-trained employees. So how would we measure that? We know that throughput is not a good metric; AI is verbose and can produce thousands of code lines of non-performant slop in minutes. So how do you measure the value of high-quality business analysis? How do you measure the value of addressing the right problem space?

These are going to become meaningful questions going forward as the cost of this kind of intelligence work becomes more salient. We’re going to have to think about the trade-offs of the costs of AI and of hiring the right person to use AI, and how to know what that combination truly buys us.

As if work isn’t difficult enough already, we are getting news and updates about AI technology and implementations at a blistering pace while we are also bombarded with messages trying to shape our thinking about AI and agents, sometimes incorporating real-world intelligence, sometimes not. It is taxing our discernment and critical thinking skills.

But everyone agrees that AI is forcing organizations to rethink the way work gets done. We have the hype cycle telling us that “AI agents will replace workers.” A lot of people tasked with operationalizing AI and agents are reporting that “AI struggles with messiness, breaks under pressure, and needs humans more than ever.”

Here’s the uncomfortable truth:

Most AI agent deployments fail within 30 days.

Agentic AI is still an immature and developing technology, but it is also failing because real business operations are chaotic, inconsistent, and filled with exceptions that require human judgment and experience.

At the same time, junior employees whose work is mostly “production”—churning out documents, research, analyses—are feeling real pressure, because modern agents are good at eliminating repetitive, predictable tasks.

So how do we reconcile this?

How can small and mid-size companies use AI without falling for unrealistic promises?

AI agents are software products, so when we think about how they’re structured and what problems the agents are designed to address, it is instructive to look at them through the lens of Conway’s Law, and how the organizational structure informs the product structure. And let’s face it: our organizational trees are messy, political, and constantly evolving. So it is with the AI agents that we’re building. But here, I outline three practical interaction models—ways teams actually work with AI today—and layer in what we now know about economic constraints, deployment failures, and talent trends.

Matthew Skelton and Manuel Pais identify four team types in their book Team Topologies…

This piece draws from amazing content from TechButMakeItReal and Nate B. Jones.

Let’s break it down.

This is where almost everyone starts: humans lead, AI copilots assist. A PM runs competitive research with ChatGPT. A designer generates directions for a wireframe. A founder drafts an investor update.

It feels great. Productivity jumps. But there’s a hidden flaw:

Your organization doesn’t get faster—only your people do.

Copilots don’t coordinate across departments. They don’t unify context or coordinate across teams. So each of the four team types might see small individual gains, but this copilot approach won’t help the overall organization; you’ll still see the same bottlenecks (e.g., incompatible data, differing alignments) moving intelligence between teams. Although stream-aligned teams may appear to get the most benefit because they tend to be more customer-facing, but this masks the messiness in the backend.

They simply accelerate whatever silo they’re plugged into as we’re seeing across companies deploying AI:

The takeaway for smaller organizations:

Copilots are cheap, safe, and useful—but they won’t meaningfully compress your discovery → design → MVP → PMF cycle. They’re helpers, not operational assets.

This is where things get interesting. Instead of copilots assisting individuals, AI agents take on recurring, domain-specific responsibilities:

These agents don’t replace humans—but they do act like digital employees inside departments.

In team topologies, this can start to build bridges across teams as the agents become more reliable and multiple teams begin to speak the same language. Enabling teams can build agents to anticipate the most common needs from the stream-aligned teams. Internally, a data pull-system can start to take shape as the team needing a tool or asset can easily pull it into their process ready-to-go, rather than needing to ask for it.

Crucially, this aligns with what the data actually shows:

Agents are most successful when their tasks are 90% predictable and domain-expert configured.

Harvey works because ex-lawyers configure its workflows. Deployment works when someone deeply understands the operational system the agent will run inside.

This model matches the emerging reality in agent deployments:

For SMBs, this is the highest ROI model today because:

✅ It speeds up real operational cycles

✅ It reduces human admin work

✅ It improves experimentation frequency

✅ It’s achievable without massive reorgs or agent infrastructure teams

But it introduces one new bottleneck:

Your human orchestrator becomes the constraint.

You need someone who can verify agent output, connect tasks across agents, and ensure alignment. This requires problem-solving skills—not production skills. And it explains why senior talent is safer in the AI transition: expertise + judgment = irreplaceable.

This is the frontier: agents stop acting like departments and start acting like product teams. Research informs design. Design triggers prototyping. Prototypes trigger validation. Experiments run continuously.

Humans set goals and constraints, agents execute.

Imagine finding new data, tool, or assets in your workflow that you didn’t know you needed. The AI itself anticipates your needs, feeds that back to the supporting team, and that supporting team can deploy an asset with minimal human interaction, and you are now smarter in the moment, rather than needing to have complicated coordination meetings.

It sounds magical—and early demo videos make it look real—but here’s the grounded reality:

The value here isn’t autonomy—it’s speed of iteration.

When the workflow is well-structured, this model can compress weeks into hours.

But it requires deep investment in:

This is not a “plug and play” model today.

But it’s where we’re headed.

When you synthesize both the agent economics and the human talent perspective, one coherent truth emerges:

AI isn’t replacing human work. It’s replacing the predictable parts of human work.

That means:

Small to mid-size teams have a unique advantage because they can:

✅ Move faster than enterprises

✅ Avoid heavy agent infrastructure

✅ Integrate agents into workflows without bureaucracy

✅ Train employees on problem-solving while roles are still flexible

Here’s the actionable guidance I’m giving every SMB leadership team right now:

1. Start with copilots to identify where your real bottlenecks are.

Let individuals explore and build internal champions.

2. Introduce functional agents in the highest-repetition parts of the org.

Research, QA, customer support, validation, documentation.

3. Require every employee to validate AI outputs.

Verification is the new basic skill.

4. Shift talent expectations from production to problem-solving.

The safest employees will be those who combine domain expertise with AI leverage.

5. Move to multi-agent workflows only when:

That’s how you avoid the 90% failure rate—and start building real compound advantage.