Bait the hook with food that fish like.
The fundamentals of getting a concept into market quickly, then learning from real users has never been more applicable than in 2026, the year where AI Engineering / Spec Engineering / Loop-based Orchestration will remake how software product companies (and their private equity investors) commercialise innovation.
I’m making the case that this chart (poached from Savoia’s book “Pretotyping”, but with its conceptual roots in The Lean Startup) is actually more true in 2026 than any of the 16 years since its launch.
It doesn’t matter if you’re a hot new AI Native startup, a 9 year old Series B scale up, or a 122 year old sandpaper company: We all have the pressure to innovate. Unchanged in 2026 is that our competition is pushing us forwards at ever-accelerating rates.
The conceit that “our fully-baked idea is the right one” will persist forever, I know because I succumb to this temptation just as much as the next person, but it’s counterproductive.
I’ll butcher Carnegie’s quote into something more concise: “Don’t bait the hook with food that you like, bait it with food that fish like”. How do we find out whether fish like the bait we’re using? Launch Early and Ugly to see if anyone likes your choice of bait.
4 years ago, I used to wince at MVP builds that cost north of $100k. In 2025 I would wince at MVP builds that cost north of $25k.
Now in the 2026 world of Spec Engineering, there’s (apart from some niche exceptions; @ me in the comments) very little justification for your early concepts not being in the hands of real users within weeks, or at most, a month or two.
In general, you (even if you’re non-technical) should start playing with the 2025 “Level 3” tools for Agentic Software Development like Claude Code and OpenAI Codex. Once you’ve got your head around these, start playing with the “Level 4” tools for Loop-based Orchestration. There are three more coming out every week, but Loom or Swarm-CLI are fun places to start.
What does this mean for you?
If you’re pre-revenue and non-technical, this doesn’t automagically mean it’s safe to launch with these vibe-coded products. In 2026 there’s still a material role for Software Engineers (well, right now there is anyway, who knows what September will bring), but this at least allows you to cohere your thoughts, create some short videos, and run some Zoom-based demos to get feedback prior to engaging a “real engineer” to wrap some guardrails around your Spec for the “Version 5” when we have more conviction that Fish Like This.
If you’re post-revenue, non-technical, and employ software engineers, you need to be talking to them about what level of AI Engineering tooling they’re currently using, and support them with resources (time) to run spikes within/across teams to get a feel for how to move up to the next level. Doesn’t matter if it’s Level 2 -> 3 or Level 3 -> 4, they just need to get going on this pronto.