One of my advisors recently told me that our website looked "ghetto."
I'm not going to lie: it hurt my feelings.
Team Ignite Ventures has spent years building a serious venture platform. We've made hundreds of investments, built an incredible network of founders and investors, and developed a differentiated approach to early-stage venture capital. We work hard to be thoughtful, responsive, and useful to the people in our ecosystem.
But if someone's first interaction with Team Ignite was our website, none of that came through.
The worst part was that my advisor was right.
Our old website was built on Squarespace using a traditional "what you see is what you get" editor. It had served its purpose, but it had become dated, difficult to evolve, and increasingly disconnected from what Team Ignite had become.
It didn't communicate the energy of the firm. It didn't make our strategy easy to understand. It didn't give founders, limited partners, co-investors, or potential partners an obvious path to what they needed. It looked like a website that had accumulated incremental edits over time, because that is exactly what it was.
The site was long overdue for an overhaul.
This summer, in the doldrums of meeting-less days, I decided to rebuild it.
The twist is that I didn't hire a design agency or a team of developers. I sat down and built it by talking to AI.
Building a website by having a conversation
The phrase "vibe coding" gets used loosely, but this project showed me what it actually looks like in practice.
I didn't sit down and write thousands of lines of code. I described what I wanted. I reviewed what the AI built. I reacted to it in plain English. Then I asked it to revise, research, test, and improve.
The workflow looked something like this:
"The portfolio page is good, but these logos don't work against a black background. Find better versions."
Then:
"The team photos are being cropped strangely. Look at the old website and match the framing more closely."
Then:
"LPs shouldn't be sent to a calendar. Build a contact form that sends us an email."
Then:
"The Syndicate form needs to work like the existing one and add submissions to our Google Sheet."
Then:
"This section feels too small. Increase the type scale, but don't change the underlying simulator."
Then:
"The header needs to show visitors which section they're currently viewing."
That was the development process: conversation, implementation, review, feedback, and another pass.
I would give the AI a task, go do something else, and come back five or ten minutes later to review the result. Sometimes the first version was excellent. Sometimes it was directionally correct but needed taste and judgment. Sometimes it misunderstood what I meant, and I had to explain it differently.
Then we would repeat the cycle.
I did this throughout the day for roughly a week to get the first version live. All in, I probably spent around 40 hours actively reviewing, directing, testing, and refining just to launch. And, as you'll see below, I haven't really stopped since.
That is still a meaningful amount of work, but what came out the other side would traditionally have required a designer, a developer, a project manager, multiple review meetings, and a much larger budget.
ChatGPT won the first round
I used both ChatGPT and Claude during the project.
Both were useful, but ChatGPT did a better job on the initial design, which was surprising.
It more quickly understood the visual direction I wanted: oversized typography, generous negative space, a near-black background, restrained use of Team Ignite's amber and orange brand colors, and a level of confidence that felt appropriate for a modern venture firm.
The first design immediately felt much closer to us than the old site did.
Claude was helpful as another set of eyes, particularly for detailed reviews, content audits, data checks, and identifying areas that needed another pass. One model might create a strong implementation, and the other could critique it. Then ChatGPT could take that critique, inspect the actual code, and implement the useful parts.
Using multiple models started to feel less like switching between software products and more like working with a small team that had different strengths.
Still, the quality of ChatGPT's initial design surprised me. It didn't merely generate a generic venture capital template. It translated a long, messy set of thoughts about our brand, audiences, portfolio, and investment philosophy into a coherent visual system.
That gave the entire project momentum.
Designing for four different audiences
One of the harder parts of the site was that Team Ignite doesn't have one audience.
The site needs to work for at least four:
- Founders looking for funding
- Limited partners learning about Team Ignite
- Venture firms and co-investors interested in our network and Syndicate
- Operators, advisors, podcast guests, and other potential partners
The easy solution would have been to cram four competing calls to action into the homepage hero. That usually produces a site that speaks to everyone and connects with no one.
Instead, we kept the homepage focused on the identity of the firm: venture capital is a team sport.
The navigation then gives each audience a clear path. Founders have a dedicated section and deck-submission flow. LPs have a page explaining what it means to be part of the Team Ignite network and a form to begin a private conversation. Co-investors can learn about the Syndicate and register their interest. Other visitors can explore our portfolio, team, resources, blog, and podcast.
That information architecture sounds obvious in retrospect. Getting it right took a surprising number of iterations.
Putting a face on the firm
If venture capital is a team sport, the team should actually be on the website.
We built out a proper Team page: real bios, real headshots, a clear sense of who's behind the decisions Team Ignite makes. It sounds like the simplest page on the whole site, and in some ways it was; no simulator, no live data feed, no forms. But it forced the same discipline as everything else. Photos that looked fine on Squarespace didn't hold up against a near-black background at full size. Bios that read fine as a paragraph on an old page felt bloated next to the rest of the site's restraint.
It's a small page doing an important job: reminding anyone who lands on the site that there are actual people building this network, not just a fund name.
Turning our portfolio into a living part of the site
The portfolio page became one of the biggest pieces of the project.
We brought in real portfolio data, sourced and cleaned company logos, added descriptions, sectors, stages, and notable co-investors, and built client-side filtering so visitors can explore the portfolio by category.
We also spent a lot of time on the details no one notices when they work, but everyone notices when they don't.
Some logos looked great on white backgrounds but disappeared on black. Others had low-resolution source files, awkward padding, or incorrect proportions. A few companies had changed names or domains. Some stage and investor information had become stale as companies raised subsequent rounds.
AI helped research, organize, and implement the updates, but this part also demonstrated why human review remains essential. A logo can be technically valid and still look terrible. A company can be categorized in a way that is defensible but unhelpful. A public funding announcement can be newer than your internal database, or the internal data can be ahead of what is public.
The finished portfolio is not just a grid of names. It is a much better expression of the companies and networks behind Team Ignite.
We also deliberately kept performance information off the public website. Venture capital operates in a regulated environment, and a marketing site is not the place for public return claims or detailed fund metrics. Those conversations and materials belong in private LP communications and the LP portal.
That compliance constraint shaped the copy throughout the site. The goal was to be compelling without drifting into general solicitation or turning the website into a public pitch deck.
The jobs board: the network, made literal
If there's one addition since launch I'm proudest of, it's this one.
We say all the time that Team Ignite's value is the network, that our portfolio companies benefit from being part of something bigger than a single check. It's easy to say that on a website. It's much more convincing to show it.
So we built a jobs board that aggregates real, live openings across the entire portfolio, automatically, every single night.
It's not a static list somebody updates by hand. There's a crawler that visits each portfolio company's own site, recognizes the ATS platform they use (Greenhouse, Lever, Ashby, Workable, and half a dozen others), and pulls in whatever's actually open, refreshed daily. For companies whose job listings only live behind a client-rendered page (increasingly common), it falls back to actually rendering the page like a real browser would, rather than giving up. And for the handful of companies our crawler genuinely can't see into on its own, we built a simple self-serve form: a founder picks their company from a list and drops in a link, and it flows straight onto the board.
The result, as of today: hundreds of open roles across dozens of hiring portfolio companies, refreshed every night, with zero manual upkeep.
That's the part that made this feel different from a typical redesign project. It's not a page. It's a live pipeline, quietly running in the background, doing real work for real founders trying to hire, every day, without anyone touching it.
The podcast and blog finally live at home
For a while, our podcast and blog existed in that slightly awkward state a lot of companies end up in: real, ongoing content, hosted somewhere else, linked to from a page that wasn't really designed for it.
Now they're properly part of the site. Podcast episodes and blog posts are pulled in dynamically instead of being hardcoded, so publishing a new episode or post doesn't require touching the website at all; it just shows up. It's a small thing structurally, but it matters: the site now actually reflects that Team Ignite is publishing and thinking out loud constantly, not just raising and deploying capital.
(Which, incidentally, is exactly how this article will get there too.)
Bringing our thinking onto the site
We didn't want the new site to be only a digital brochure.
One of the most interesting additions is our Volume Thesis simulator, which helps visitors explore the relationship between portfolio construction, selection skill, reserves, ownership, and venture outcomes.
The simulator previously lived as a separate tool. We integrated it into the website under Resources and restyled its presentation so it feels like part of the same experience, while being careful not to change the model's underlying content or logic.
This is where the conversational nature of AI development became especially powerful.
I could say:
"The simulator works, but everything feels too small."
Then review the updated typography and spacing.
Later, I could say:
"When we scroll, only the Global Settings section should stay locked above the reserves settings."
The AI could inspect the existing implementation, make the change, and let me test it immediately.
That kind of iterative product conversation used to require me to translate what I wanted into a formal ticket, explain it to a developer, wait for implementation, and then schedule another review. Here, the feedback loop was measured in minutes.
Forms that actually do something
A polished website is not very useful if the forms disappear into a black hole.
We built a general contact form that sends inquiries to our team by email. We connected the Syndicate form to the Google Sheet used by our existing process. We added validation, submission states, error handling, and protections against repeated submissions.
Then we tested the forms with real entries in both local development and production.
This is another area where AI changed the experience. It wasn't limited to writing frontend code. It helped configure the email service, connect the Google Sheets credentials, diagnose a production credential-formatting problem, update the deployment environment, redeploy the site, and verify that the submission reached the confirmation page.
I was able to work through infrastructure and deployment tasks by talking to the AI in the same way I talked to it about typography.
That is wild.
From localhost to a real production deployment
The AI also helped create a private GitHub repository and introduce proper revision history.
That matters because vibe coding can otherwise become dangerously informal. If every iteration exists only on one computer, a great experiment can turn into a fragile mess.
We committed stable versions of the site as we went, pushed the release to GitHub, connected the repository to Vercel, configured production environment variables, and connected the new teamignite.vc domain through Namecheap.
The AI helped trace the entire chain:
- Local development
- Source control
- Private GitHub repository
- Production build
- Vercel deployment
- Environment variables
- DNS records
- SSL
- Custom domain
- Production form testing
At the end, both teamignite.vc and www.teamignite.vc were serving the new site.
A week earlier, this was an aging Squarespace site. Now it is a responsive Next.js application with typed local data, a reusable component system, working backend integrations, SEO metadata, a sitemap, a podcast feed, a blog, portfolio filtering, an interactive venture portfolio simulator, a real team page, and, as of the last few hours, a jobs board that runs itself.
AI makes building easier, but judgment still matters
This experience didn't convince me that human judgment is becoming irrelevant. It convinced me of almost the opposite.
AI makes execution dramatically cheaper and faster. That increases the importance of knowing what you want.
The AI could build a portfolio card, but I still had to decide whether the information on it was accurate and appropriate.
It could find a logo, but I had to say when the logo looked wrong.
It could write LP copy, but I had to determine whether the tone was too promotional for a regulated offering context.
It could design a page, but I had to react to whether it felt like Team Ignite.
It could crawl hundreds of companies' career pages looking for open roles, but I had to decide how thorough was thorough enough, and where the line was on what it shouldn't try to scrape.
It could implement almost any individual request, but I had to keep the entire project coherent.
The work shifts from manually producing every artifact to directing, evaluating, and refining. Taste becomes more important, not less. So does the ability to articulate why something feels wrong.
The funny thing is that I didn't begin the project with a perfectly formed specification. I discovered what I wanted by reacting to what the AI created.
That is part of the magic.
The first version gives you something concrete to respond to. Your response helps create the second version. By the fifth or tenth iteration, you have made dozens of decisions that would have been difficult to express upfront in a traditional requirements document.
The specification emerges through the conversation.
We are living in impressive times
I've built technology companies. I invest in technology companies. I spend most of my time around founders building at the edge of what is possible.
Even with that context, this experience felt like a glimpse of a genuinely different future.
I could sit at my computer, speak naturally, and build something of remarkably high quality.
Not a mockup. Not a prototype that falls apart when someone clicks it. A real website, connected to real systems, running on our real domain. And, months later, still growing new capabilities instead of sitting frozen the moment it launched.
The distance between an idea and a working product has collapsed.
That doesn't mean everything happens instantly. It doesn't mean the AI is always right. It certainly doesn't mean you can stop paying attention. I spent around 40 hours on the initial launch because I cared about the details and kept pushing it to improve, and I've kept going since, because it turns out the same conversation that built the site can keep building on it.
But those hours produced an outcome that would previously have taken far more time, money, and coordination.
And I enjoyed it.
There was something addictive about having an idea, describing it, and seeing it become real minutes later. It felt less like operating software and more like collaborating with an endlessly patient creative and technical team.
My advisor's comment hurt, but it gave me the push I needed.
The old site no longer represented Team Ignite. The new one does, and it's still getting better.
Take a look at teamignite.vc.
And if something still looks ghetto, please be slightly gentler when you tell me.