Five Levels of AI Building, From Idea to Market

Every week, a new model drops. GPT-something-larger. Claude-something-faster. And somewhere, a builder who spent three months shipping a product watches the announcement and wonders if they just got erased.

Nate B Jones, who runs the AI News & Strategy Daily channel and newsletter, thinks that fear is real—but mostly diagnostic. In a recent video, he laid out what he's calling the five levels of AI building: a framework for understanding not just what you're building, but how you're thinking about the act of building itself. His argument, stripped of the motivational scaffolding, is this: when a lab launch panics you, that's useful data about where you actually are.

It's a sharper provocation than it sounds.

Level One Is Not an Insult. It's a Starting Point.

Jones describes a level one builder as someone "passionate about what AI can do for the space and passionate about their particular idea"—and that's it. No go-to-market thinking, no framework for the problem space, no thesis about how the market will evolve. Just the idea, and the electricity of it.

This isn't a character flaw; it's a phase. The problem Jones identifies is that level one builders are structurally fragile. Their entire bet is that their idea is correct and that no one else—including a well-funded frontier lab—will touch their specific corner of the market. That's a narrow bridge to walk across.

The move to level two is deceptively simple: talk to customers, and let what you hear actually change what you build. Jones points to a CRM builder he encountered who had genuine domain passion but was willing to interview ten customers and adjust. "That is already so far down the road in building a business," he says. The product idea stays recognizable; the specifics shift to meet reality. Jones claims to have seen this level of insight produce five- and six-figure side businesses—modest by venture standards, meaningful in absolute terms.

The Go-to-Market Gap Nobody Wants to Talk About

Level three is where Jones starts threading something genuinely new into otherwise standard startup advice. The classic piece—you need distribution, you need to understand how to reach customers, storytelling matters as much as product—has been gospel in startup circles since at least the Steve Blank era. What Jones adds is that AI has changed the economics of go-to-market in ways builders aren't fully internalizing.

The tools he cites are specific: AI-driven LinkedIn outreach with custom messaging, Twilio plus a voice model for customer calls, HeyGen for video storytelling, model-driven TikTok accounts. None of these are hypothetical. The point isn't which tool you pick—it's that a small team can now generate and distribute narrative at a scale that used to require a marketing department. "AI businesses, AI startups are actually leveraging this technology all the way across all the functions in the business," Jones says. The builders who treat AI as purely a product ingredient, and not as infrastructure for sales and distribution, are leaving a compounding advantage on the table.

This connects to a broader tension in the AI entrepreneurship shift that's been reshaping who can realistically build a business: the cost of customer acquisition is falling fast for people who know how to use these tools, while rising in confusion for those who don't. Level three is where that gap starts to matter.

What a "Thesis" Actually Means at Level Four

Jones's level four concept is where the framework gets philosophically interesting, and a little harder to package as actionable advice. A level four builder has what Jones calls "a deeply uncomfortable but strongly held conviction about the world"—a thesis about their problem space that doesn't wobble when a new model releases.

He reaches for voice computing as his illustrative example, and it's a good one. Wispr Flow (wisprflow.ai), a voice-to-text tool Jones uses personally, exemplifies his argument: the team didn't just build a transcription product. They oriented their entire product architecture around the conviction that voice is the next computing paradigm. That means obsessing over capture quality, hotkey accessibility, seamless integration across applications—every detail flows from the foundational belief. It's not a feature set. It's a worldview expressed in software.

"No lab is going to be able to spend as much time in this domain as I have," Jones argues, framing deep domain knowledge as an inherent advantage over frontier labs that necessarily spread attention across everything. The labs set the capability floor; the specialist builder knows the ceiling of what the market can actually absorb and where the real friction lives.

This is the structural argument for why OpenAI and Anthropic aren't simply vacuuming up every possible business opportunity. The counterargument worth sitting with: labs have absorbed entire product categories before—witness the fate of standalone grammar checkers after GPT-4 landed, or the compression of the AI writing assistant market after Claude and ChatGPT added long-form generation to their base offerings. The thesis-based builder Jones describes is more resilient than a features-based one, but resilience isn't immunity. The question of which domains are truly lab-proof versus temporarily protected is one Jones gestures toward but doesn't fully resolve—probably because nobody can.

This tension is part of why the AI business models question remains genuinely open heading into the next cycle.

Level Five Is About Time, Not Intelligence

The final level Jones describes is the most interesting to think about—and the hardest to fake. A level five builder doesn't just understand what AI can do now. They understand the trajectory of AI in their specific domain, well enough to build for capabilities that don't exist yet.

"You are going to be able to build a business that focuses on the emerging AI capabilities that will be possible in six months or twelve months in your problem space that aren't possible today," Jones says. The mechanism is actually legible: you follow what the labs are releasing, you understand the current capability envelope (context windows, tool calling, agentic session length), and you translate the directional trend into specific implications for your domain. If you're deep in outbound sales, you can reason about what better multi-step agentic workflows will make possible for prospecting in a year. If you're in voice computing, you can anticipate what lower latency and better speaker identification will unlock for enterprise use cases.

What distinguishes this from ordinary trend-watching is domain specificity. Jones is careful about this: "There are a lot of people who will do straight-line extrapolations from general news, but there are fewer people who know their individual domains well enough to have unique insights about their particular domains." Being first to recognize that agentic sessions are getting longer is easy. Being first to understand exactly what that means for compliance workflows in mid-market healthcare is something a generalist can't fake. That's the actual moat.

It's also worth noting what this framework demands: not just technical literacy, but the kind of accumulated domain knowledge that tends to come from years inside a specific industry. The non-technical builder moment has real teeth, but level five arguably belongs to people who have that plus deep functional expertise—a combination that's less common than the current enthusiasm for "anyone can build with AI" suggests.

The Diagnostic Value of the Framework

What's most useful about Jones's five levels isn't the roadmap—it's the diagnostic function. When a new model announcement makes your stomach drop, that feeling is telling you something. It might be telling you that your competitive position is genuinely thin. Or it might be telling you that you haven't yet articulated why your thesis holds regardless of what the labs ship next.

Those are different problems. The first one requires a strategic pivot. The second one requires you to sit down and do the hard thinking—about customers, distribution, domain conviction, and where AI in your specific world is actually headed.

Jones puts the transition from level three to level four plainly: "That is a bit of a chasm. That is a bit harder." He's right that this is the hardest jump in the sequence. Customer feedback and go-to-market are learnable skills with established playbooks. Developing a genuinely disruptive thesis about a problem space—one that survives contact with reality and doesn't dissolve every time Claude drops a new capability—is closer to the kind of insight that can't be scheduled.

The framework doesn't resolve the real anxiety builders are carrying. It just tries to redirect it somewhere productive: not will the labs eat my lunch? but do I understand my domain and my customer well enough that it wouldn't matter if they tried?

That's a harder question. It's also the right one.

Marcus Chen-Ramirez is a senior technology correspondent for Buzzrag, covering AI, software development, and the intersection of technology and society.