There is a well-known assembly line inside every software company. A product manager writes a requirements document. Engineers assess whether the thing is technically possible. Designers draw mockups. Engineers rebuild the feature in actual code. The handoffs are slow, the context gets lost at each station, and the whole trip can take two to three months. A San Francisco startup called Lightsprint has now raised $6.1 million on the observation that the middle two stations might be optional, as Shubhangi Goel reported at Business Insider on Monday.
The mechanism is worth stating precisely, because it is the interesting part. A product manager, designer or operations person types a plain-English request. Lightsprint’s AI agents, connected to the company’s existing codebase, can write the product requirements document, generate the code and produce a live preview. The output is not deployed software; it is a pull request — a proposed code change that a human engineer then inspects before anything is merged into production. The non-engineer gets to ship, in a specific technical sense that stops one step short of shipping.
“We found that actually there were a lot of PMs like me as well as designers who actually really wanted to shift to production,” Benjamin Ong, the cofounder and CEO, told Business Insider. The demand side of this is easy to see. Designers already use generative AI for early concepts and prototypes; product managers have taken to agents and vibe-coding tools; and the US Bureau of Labor Statistics projects 5% growth in web and digital interface designer employment from 2025 to 2035, which suggests the designers are not leaving so much as getting new tools.
The company — founded earlier this year by Ong, Benedict Chan and Heng Hong Lee, and fresh out of Y Combinator’s Spring 2026 batch — is positioning itself against the fragmented toolkit companies currently use: prototyping software like Lovable, design software like Figma, coding assistants like Anthropic’s Claude. The pitch is that Lightsprint is a one-stop shop over the existing codebase rather than another app in the pile. The phrase Ong uses for what it replaces is that engineers, product managers and designers still work together in a way that is slow and, in the demo description, janky.
The commercial details: the seed round was led by 645 Ventures, with January Capital, Rebel Fund and angels including Opendoor CEO Kaz Nejatian. Lightsprint began with early users in April and says it now works with over 100 companies. It runs a six-person team split between San Francisco and Singapore, and the new money goes to hiring across business sales, marketing, design and engineering.
Now, there is an obvious way to read this product, which is ‘let people who can’t code commit code,’ and a more accurate way, which is ‘move the bottleneck one step down the hall.’ Someone still has to review the pull request. Ong says as much: “We’re seeing product managers, designers, and even business teams build features and submit high-quality code changes themselves, while engineering stays in control.” That is the entire business model in one sentence, and also its entire tension. The chain of handoffs gets shorter; the queue at the last door gets longer. A recent study of hundreds of firms found that AI coding agents added code faster than anyone could review it, with review time climbing even as output barely moved. Lightsprint’s plan, in effect, is to feed more pull requests into that queue — from people whose job descriptions never used to include writing code at all.
None of which is to say it won’t work. Two to three months of handoffs compressing into an afternoon of agent-generated diffs is a genuinely good deal, provided the reviewing engineers keep reading. The pull request is doing a lot of work here: it is the UI, the contract and the alibi. “Engineering stays in control” is a comfort right up until the day engineering is reviewing four hundred pull requests a week from colleagues who, by design, could not have written any of them.