Engineering leaders who ignore GenAI do so at their peril. Founders who don’t adopt it will burn far more cash and ship much more slowly than those who do. Engineering velocity is compounding week by week, and teams that lag risk being left behind entirely. With every developer equipped with a superpowered copilot, teams are making massive efficiency gains. According to the Nielsen-Norman Group, coding assistants already enable developers to complete 126% more projects per week, and these gains are expected to grow as the technology and workflows mature.
But engineers can’t just hand GenAI free rein over the codebase. The same models that turbo-charge velocity can hallucinate edge cases, skip tests, or miss the nuances of production scale. Human expertise is crucial for guiding and contextualizing AI to achieve its maximum impact.
Every engineering team must strike a balance between harnessing GenAI’s productivity boost and enforcing robust guardrails. Day-to-day workflows will vary depending on the company's stage and the team's needs. To see how this balance plays out in practice, we sat down with two RSCM founders at different stages of startup maturity to learn their real-world approaches to GenAI and engineering.
Founder Bios and Engineering Focus
Kristina Fan is the CEO and Co-Founder of 7-Chord, an AI pricing and analytics firm that provides solutions in the $150 trillion fixed income market. BondDroid AI, 7 Chord’s flagship product, is a proprietary AI agent that empowers traders and investors worldwide with timely market insights 24 hours / 5 days a week. Kristina is a technical founder who started the company in 2017 and leads a lean engineering team. Together, they work on optimizing and further enhancing an established product that has already achieved product-market fit.
Nick Ross is the CTO and Co-Founder of Home Spritz, a Canadian marketplace that vets professional cleaners and places them with homes and businesses. Nick is the only engineer on the team, and as Home Spritz expands its market, Nick’s primary concern is shipping faster and increasing his output. He uses AI not just to write code, but to explore ideas, debug logic, and maintain momentum.
How AI Fits Into the Tech Stack
AI coding assistants are a lifesaver for a solo engineer like Nick. “I want to highlight that Home Spritz is still in an early stage of growth. We're still figuring things out. So for us, iteration is king,” Nick explains. ”Seeing a tool that could give us an advantage in terms of velocity has always been a no-brainer.”
“We’ve used a range of AI tools—from ChatGPT to other models—to generate code, refine it, and integrate it into our codebase. We began with Copilot and have since adopted Cursor as our primary development environment,” says Nick. “To make AI truly effective, we’ve found that using a unified toolchain is key: Our entire stack runs on TypeScript and JavaScript, from the backend and REST APIs to the React / Next.js frontend. As for the rest of our stack, we’re on AWS, with MySQL as the primary database and Redis for caching.”
Kristina’s team approaches AI differently and has distinct use cases for augmenting their work. “In contrast to Nick’s company, we are far along. So most of our uses for AI are not necessarily in iteration but refactoring,” she explains. “We use VS Code plug-ins such as Gemini Code Assist and Cursor extensively for testing, refactoring, and optimization, and ChatGPT Plus for code review and documentation. Because our product is a mature, back-end-only API, the bulk of work we outsource to AI is on maintaining and optimizing existing code rather than building new front-end features.”
When it comes to tech stack, Kristina’s team runs human-designed proprietary real-time AI inference and streaming pipelines on Google Cloud, built around Postgres (Cloud SQL), MongoDB, RabbitMQ, and Redis, with Cloud SQL as their only managed service.
AI as Code Reviewer
Kristina has turned generative models into an extra set of eyes on every pull request. “We have automated a lot of the code review,” she explains. “We use detailed, templated prompts that surface common errors and deviations from our code style guide.” The key to prompt success is blunt questioning. Kristina shares how vague prompts lead to flattery, while direct ones surface actual issues. “For example, when you ask, ‘What do you think about this code?’ it tells you it’s great. But if you ask, ‘What’s wrong with this code?’ then it gets honest.”
Those AI comments form only a first pass. “None of the AI tools are as good as traditional linters like Pylint for Python or automated security scanners,” says Kristina. Their team members often run several models side by side, compare their sometimes conflicting flags, and then make the final call themselves. “AI tools are still not effective at identifying significant architectural flaws, so I don’t see a way to eradicate human code review.”
Nick uses AI reviewers to decode logic just as much as he uses it to police style. “If I’m exploring a library and I come across a piece of code that I don’t understand, AI can help me make sense of it,” says Nick. “It's insane how good it is as a teaching tool, even for a developer like myself with a decade of experience.”
Recently, the same approach helped him untangle a four-year-old block of his own code. “I was like, who did this? Why would the person do this?” Nick laughs. “The person, of course, is me. But AI actually did help me understand and rewrite a piece of code that I myself wrote years ago.”
The Cost of Blind Trust
Nick warns that every autogenerated snippet must be questioned from the outset. “You have to approach AI-generated code with the mindset that it will probably be wrong. You cannot blindly trust it,” he says, noting that subtle breakages may surface days later. The risk compounds when code touches live data. “Handing an agent SQL‐write privileges is extremely dangerous because you don’t know when it might drop or rewrite something it decides is not to its liking.”
Nick shares an anecdote about a sticky booking bar on Home Spritz’s site, a small UX feature that drives 30% more conversions. AI tools keep removing it during unrelated code changes. “It doesn’t understand what’s important to the business,” he says. Nick makes sure to review everything, test hard, and never grant autonomous write access because each unchecked error costs time and, potentially, revenue.
Kristina strikes the same note of vigilance, calling AI output merely a starting point. “By no means does any AI generate production-quality code,” she says. “We still absolutely have to review it and test it with our eyes.” She explains that models can skip lines, repeat the same mistake, or ignore standards that any linter would flag.
“These models were probably trained on bad code, and these errors just propagate,” she says. Until models learn to adhere to basic style and security rules on their own, Kristina insists that a solid engineering background and a skeptical code review remain irreplaceable safeguards.
Hiring in the Age of AI
AI is reshaping org charts as quickly as it’s reshaping codebases. Kristina revised her hiring plan after observing AI automate routine documentation. “A technical writer is permanently off the hiring plan,” she says.
AI tools also reduce the need to hire junior engineers. Instead of scaling headcount as the product matures, Kristina now hires selectively for specialized expertise. “Fixed Income quant engineers who also have domain knowledge bring deep understanding ot the business logic and customer needs to the table, besides knowing how to write code. ML/AI engineers add value with architecture and design,” she says. “AI tools replace somebody who just implements detailed product or technical requirements.”
Nick sees the same shift at the internship level. Last year, the company had a human intern, but today, he admits, “I’m not sure that’s needed anymore because the experience with AI is essentially similar to having a couple of interns.”
“AI agents are really eager to do the work and mostly capable,” he says. He concedes that he must be ready to correct mistakes, but he feels that an intern would require the same level of oversight. “In practice, the experience is identical to having interns or maybe junior people on your team,” he adds.
What’s changing isn’t just the who, but also the when: GenAI is reshaping both the cadence and criteria of engineering hiring. With AI in the loop, mid-level and senior engineers can accomplish significantly more, reducing the need to scale teams as aggressively as in the pre-AI era. Startups can now hire at a more deliberate pace, stretching every dollar and maximizing the productivity of existing team members. Headcount growth is shifting from focusing on raw bandwidth to prioritizing the addition of specialists who can think critically and make informed decisions, thereby having a direct impact on the business.
The New Definition of Engineer
GenAI has already proven it can double weekly output, but the founders who perform best will balance their speed with discipline. Kristina’s mature bond pricing platform and Nick’s scrappy marketplace sit at different stages of the startup arc. Yet, both founders draw the same line: AI is a prolific junior partner, not an infallible senior engineer. That distinction reshapes everything from daily pull-request etiquette to strategic head-count math.
Developers have entered a new era. GenAI isn’t just rewriting code; it’s rewriting what it means to be an engineer. As amazing as GenAI’s gains may be, the companies that thrive will be those that let the machines sprint, but keep the humans firmly at the steering wheel.
