There has never been a better time to take a swing at starting a company. What needed a six person team and two hundred thousand dollars in 2015 can now be built from a kitchen table. People who would never have been able to try are trying, and that is genuinely great. At Strongly we love watching someone chase the thing they have been thinking about for a decade. This post is for them. It is not meant to talk anyone out of anything. Start the company. The point is to be honest about where you stand so you know what to sharpen and who to bring on. Self-awareness is cheaper than two years of burned savings.
The tension is that the same tools that let more people swing also let them skip the work. AI does not just lower the barrier to building, it lowers the barrier to feeling like you built something. That is a different problem, and I keep meeting the casualty of it at events. Industry veteran, fifteen plus years in some vertical, no technical background, trying to vibe code an AI company themselves on nights and weekends. No cofounder, no CTO, no engineering hire. Just them, Cursor, and a Claude tab open. They are funding this with savings and sweat equity. They cannot evaluate what they are shipping. They cannot tell whether the code their AI assistant just generated is production grade or held together with string. They cannot read a stack trace, cannot reason about why a model output drifted yesterday, cannot tell whether their architecture will fall over at ten users or ten thousand. They are six months in, quietly drowning, and convinced they are almost there.
“This is dot-com cosplay in 2026, and we are early in the cycle.
But the vibe coded unicorn stories say otherwise
You have heard them. A solo founder ships in six months and sells for eighty million. A bootstrapped app hits a hundred million in ARR before its first birthday. YC has been preaching ship fast for two decades. Lean startup has been gospel for fifteen years. The new AI version of the gospel says you do not even need to be technical anymore. Vibe code it, get it in front of users, iterate.
Some founders are pulling this off. The published profiles also flatten the picture. Maor Shlomo sold Base44 to Wix for eighty million in cash six months after launch. He had also co-founded Explorium, an Insight Partners backed analytics company that raised over a hundred million dollars, served in the Israeli Intelligence Corps, made Forbes Israel 30 under 30, and hired eight employees in the month before the sale. The solo founder vibe coder framing in the headlines was technically accurate and substantively misleading. He was a seasoned technical founder running his second company, not a non-coder discovering AI. Josh Mohrer vibe coded Wave AI to seven million in ARR alone, after years as one of Uber's first forty employees and the General Manager who scaled New York operations to three billion dollars in annual fares. The rest of the headline names (Lovable, Replit, Anything, Cursor) are developer tools and consumer apps where users churn quietly and a bad version costs you someone you never met. The iteration loop is cheap.
That is not your situation. You are a non-technical industry veteran building enterprise software for a vertical where every prospect is someone you know personally. The lean playbook assumes you can evaluate what you ship, that your buyers churn anonymously, and that bad versions cost you replaceable users. None of those assumptions hold for you. Iteration is not free. Each pitch is a relationship. You cannot debug what you do not understand, so you cannot iterate your way to a product you cannot evaluate.
Ship fast still applies. Lean still applies. The lean version of your situation is not "ship a half built product to your former CEO." It is fifty discovery conversations, one specific painful workflow, validated willingness to pay, and a technical cofounder, before you ever pitch.
Two axes will tell you how much risk you are walking into
Where you sit on each axis decides what you have to sharpen and who you have to bring on. Most founders score themselves generously on the half they are weaker in. Be ruthlessly honest with where you land on the picture below.
Can you actually evaluate the AI work?
AI specifically, not just code. Plenty of strong engineers cannot tell a real model from a demo. From most exposed to least.
You speak in outcomes and KPIs. You cannot tell whether the eval suite your team built is rigorous or theater. You cannot read a benchmark and know which numbers are load bearing. You cannot evaluate a build versus buy versus fine-tune decision. You will overpay for compute, underspend on evals, and discover at the worst possible moment that your "proprietary AI" is a thin prompt over an API your competitor uses too.
You have shipped features near AI. You can read a spec, you have sat through architecture reviews, you know what RAG and a vector database do in the abstract. You can hire without getting fleeced. You still cannot tell whether the model regression last week is a data issue, a prompt issue, or a vendor silently changing weights on you.
You shipped code in a previous life. You understand systems, latency, what a reasonable inference bill looks like. You can prototype against an API. Your blind spot is usually assuming ML is just software with more matrices. It is not. Eval design, data quality, and model behavior under distribution shift will surprise you in ways your CS degree did not cover.
Someone who has actually trained, fine-tuned, and broken models in production. You know what the frontier can and cannot do this quarter, not what Twitter said it could do last quarter. You know which problems are one prompt away from solved and which are five years away. Lowest technical risk in this cycle, and the highest leverage.
Can you actually sell AI to a real buyer?
Building is the cheap half. AI buyers right now are exhausted, skeptical, and burned. They have sat through three hundred pitches that all sound the same. Procurement has new AI policies. Security has new questions you have never been asked before. Your buyer's last AI pilot probably failed and they have not forgotten.
You have never closed a deal, never written a cold email that worked, never run a renewal. You think LinkedIn thought leadership is distribution. You will mistake hype-cycle inbound for product market fit. When that inbound dries up in eight months, you will be lost.
Marketing, BD, customer success, partnerships. You understand pipeline in theory. You have not personally carried a number. You will probably underprice and overpromise.
You have closed contracts. You can read a deal and tell when it is real versus a polite stall. You will still have to relearn parts of the motion because the AI sales cycle has its own pathologies: longer security review, fuzzy ROI conversations, pilot purgatory.
You have done this before in a category that required behavior change, not just a budget line swap. You know how to handle the "how is this different from ChatGPT" question on the third call.
The sweet spot does not exist in one person
Almost nobody is high on both axes, which means even the most technically-savvy founders carry a high commercial risk, and vice versa. Founders who think they are usually score themselves generously on the half they are weaker in. If you are a brilliant ML engineer who has never sold anything, you do not need to learn sales fast, you need a cofounder who already knows it. If you are a sales killer who cannot evaluate model work, the situation is identical in reverse, and arguably worse, because vendors will smell it on you.
“Two cofounders covering both axes will outrun a brilliant solo founder almost every time. Three is fine when the third role is real. Five is a committee dressed up as a cap table.
The cost nobody is pricing in
The savings you burn are the easy part. The hard part is what AI hype does to your judgment, and what your judgment does to your network.
The pattern is familiar by now. You see your AI assistant generate something that looks impressive. You run a prototype, watch it work once, and convince yourself you have a category defining product in a vertical you have spent fifteen years in. You start emailing the network. The CEO you used to report to. The head of ops at your old competitor. The friend who runs procurement at the biggest buyer in the industry. They take the call because of the relationship, not because of your product. They have no reason to be skeptical of you yet.
Then you show them something hollow. A demo that works once. A workflow that falls apart the moment they bring real data. A pricing slide for capabilities you have not built. They are polite. They do not buy. They also do not refer you. Quietly, they tell two other people in the industry that you have lost the plot.
That is the move that ends careers. Savings can be rebuilt in two years. Industry credibility takes fifteen to build and an afternoon to lose. The cruel thing about AI right now is how easy it is to assemble a demo that fools you about your own progress, and how expensive it is to use someone else's trust to validate that delusion. The relationships were the real asset all along. Do not spend them on something you cannot stand behind.
The hot take
Most AI companies being started right now are not companies. They are wrappers built over a few weekends, with a landing page, a Stripe link, and a vibes-based moat. The dot-com analogy is not that the technology is fake. It was not fake then either, and the survivors of that era ate the economy. The analogy is that the cost of looking like a company has collapsed. Anyone can ship a slick demo in a weekend. The bar to look credible is now lower than the bar to be credible, and the gap between those two is where the next four years of failures will happen.
“Know which side of the gap you are on before you sink another six months of nights and weekends. Then go build anyway.
If you have done the self-assessment and are still committed, here are five non-negotiable rules to follow.
Building it anyway. Five rules.
Find a cofounder in 90 days or accept your fate.
The longer you grind alone, the more attached you get to your equity, and the more your judgment about whether you need help quietly degrades. After six months you will have convinced yourself you do not need anyone. You do.
Set a hard deadline. If it passes without the right person, accept that you are building a lifestyle business and price the company accordingly. Both are valid outcomes. The trap is the founder who never made the call and ends up halfway between the two.
Use your network for discovery, not pitching.
Your unfair advantage is fifteen years of vertical relationships. The relationships are the asset, not a one-time sales channel for a demo you cannot back up. Thirty discovery calls before you write more code, show up curious, not selling.
If you cannot get a single signed LOI from your own network after dozens of real conversations, you have a hobby with a logo. If you torch the relationship pitching vapor, you have less than that.
Stop vibe coding what you cannot evaluate.
You do not need to become an engineer. You need enough literacy to know whether what you are shipping is real. Three to six months on evals, basic system design, and how to read a model card will change how you operate. Without that you are a passenger in your own company, and your AI assistant is the driver.
To be clear, this is literacy for evaluation, not capability for building. You will not learn real machine learning in a few months of prompting an LLM. People spend years mastering it, and the mathematical rigor has no shortcuts.
Pick a wedge only you can build, then check whether you actually can.
Wrappers fail because anyone can ship them in a weekend. Your real moat is vertical knowledge a generalist lacks: the workflow they cannot see, the data they cannot access, the buyer they cannot reach. Build something a smart twenty-two year old in San Francisco literally cannot ship because they have never met your customer.
Domain expertise is necessary, often not sufficient. The retired FinServ exec who knows credit scoring deeply (regulation, buyer politics, sales motion) is still not the person to build a new scoring product. Scores are ML models on complex data feeds, and that work takes a decade of mathematical rigor, not a few months of prompting. Same for fraud detection, clinical prediction, algorithmic trading, drug discovery. Two kinds of wedge exist: workflow shaped, where domain knowledge plus a strong engineer is enough, and algorithmically deep, where the product is the model and you need a real ML cofounder with years in the subdomain. Mistaking the second for the first is one of the most expensive errors in AI startups right now.
Set a kill date for your savings.
Pick a number now and write it down somewhere you cannot edit easily. The day your account hits it, you stop, regardless of how close you feel.
Founders who skip this step rationalize their way through massive token bills, ruined finances and broken marriages. The discipline of a real deadline is the difference between a brave attempt and a slow disaster.
So you read this through and you are still in
Good. If you are going to build, consider doing it on a platform like Strongly.AI. We are a team of technical operators with decades of production AI and ML experience across Fortune 500, government, and SMB customers, which is to say we have already made most of the expensive mistakes the post above warns about.
Strongly provides the plumbing that gets you roughly eighty percent of the way there, specifically addressing the technical literacy required by Rule 3. The platform delivers the security posture your buyer's CISO will interrogate, the cost controls that decide whether your unit economics survive real volume, the governance and audit trail regulated customers will require before they sign, and the observability that tells you when a model has quietly drifted on you.
This foundation lets you focus on your market wedge. Our team is there to support the journey, not replace your judgment about your own market. You still need to do the work this post describes, but the technical foundation does not have to be something you invent alone.
Build on a Foundation, Not a Promise.
Skip the eighteen months of platform mistakes. See how Strongly.AI gives founders the technical foundation to focus on the wedge only they can build.
Talk to Our Team