Smooth sailing into brick walls
The hype cycle would suggest that software development is on a fast track to being commoditized, and that senior engineers and designers are replaceable with AI / junior counterparts. First impressions with using AI app builders and agentic coding tools feel like a confirmation of this — the output is incredible. But when the natural nuance and messiness of bringing a product to market materializes, the efficacy of these tools fades into challenges. We feel strongly that building products without experienced builders leads to bad outcomes.
Quality: The persistent proof of concept
We're at the point where AI can spin up fully functioning demos without much human intervention beyond a detailed spec. But with that speed comes a drastic shift in quality, especially in the nuances and edge cases that are key to usable and delightful products.
Both the foundations (code and abstraction quality, security, performance) and experience (edge cases and non-happy-paths) are problematic without a human-driven feedback loop.
This is okay for a proof of concept. But we've seen teams treat these outputs as end-state product, ready to call it a day and move onto the next roadmap item. However, those edge cases, and the code foundation are really important - and if left unaddressed will lead to lack of adoption, churn, and loss of velocity.
Time to market: The brick wall
At a certain point of stitching together these proof of concepts, non-technical builders relying solely on AI tooling run into an inevitable wall.
It's not rare that builders will find a market foothold with a minimally marketable product generated by AI. The edge cases are overlooked, because they are solving a real pain point. But when they learn from users, iterate onfrom those learnings, and add on more and more, they are increasing the complexity. And at a certain point, an LLM stops being effective entirely.
LLMs are, by nature, good at things that they know -- things that are statistically relevant. But anyone who's brought a complex product to market knows...it's messy. No one atomic component is that complex. It's when you glue hundreds of them together in a (hopefully) cohesive and comprehensible system, that deeper experience and human reasoning is required.
Recovering from the brick wall means focused and skilled re-work. And increasing time to market in the early days to focus on the fabric can be really painful.
Momentum: The debt crisis
Given the hype, founders and teams are expecting to do more with less. We've seen teams staffed with junior engineers, expected to perform at senior levels. When coupled with AI, this can feel good during the period of a product's nascency, for the reasons described above. But it's a matter of time before the debt piles up.
Granted, this has always been the case. But AI is pushing out this realization / decision, due to the speed at which code can be written and the shortcuts you can take with an agent working for you. And pushing out the point of debt pay-off only makes it harder to deal with, when the time comes.
The momentum loss of finally having to address the spaghetti on the wall is significant.
This isn’t an argument against using these tools, but an urging to use them wisely and in the right contexts. We’re stoked about what they unlock -
Designers as engineers
Equipping a designer with the ability to build, iterate, and user-test an idea is powerful. Prototypes are a huge part of our design and discovery process. All of our designers have Cursor in their toolbelt, effectively eliminating the need for an engineer to be assisting with the prototype process. This tightens feedback loops, reduces cost of discovery, and allows engineering to focus on the messy bits.
Empowering the non-technical
We've seen on many occasions non-technical folks self-serving analytics with generated SQL, effectively iterating on user-facing details of a product via code, and fleshing out their own ideas into prototypes. This is incredibly net-positive, when the outputs are integrated thoughtfully.
Increasing engineering velocity and focus
Founding-level engineers with the ability to automate the well-defined, monotonous, and rote portions of writing code is a huge unlock, as it increases the speed (and decreases the cost) of product development, and allows for more full focus on the nuance and complexity. This requires tight feedback loops, hard-won opinions, and oversight to avoid the pitfalls above.
This landscape is changing so rapidly. The pitfalls will shift and shrink, and the benefits will grow as models (and feedback-loop driven tooling) improve. But we feel strongly that building products that humans want to use is, and will always be, a form of art.