In 2025, the question isn’t whether Machine Learning teams can work remotely, it’s whether they should.
San Francisco startups are still figuring that out. Some founders are doubling down on local, in-person teams. Others are embracing fully distributed models, hiring engineers across time zones and treating async as a default. Most are somewhere in the middle – remote-first in theory, hybrid-ish in reality, unsure what their hiring posture says about the company they’re building.
So what’s the right approach?
There isn’t one. But if you’re an AI startup hiring in San Francisco, the trade-offs between remote and on-site ML hiring are real – and your decision will shape who you hire, how you collaborate, and how fast you move.
What are the benefits of hiring remote?
The case for remote Machine Learning hiring is simple: access. When you’re not limited to a 15-mile radius around South Park, you suddenly open the door to exceptional talent, especially in a field like Machine Learning, where the top 1% may not live anywhere near the Bay.
Remote-first teams often move faster in the early stages. They hire from a broader pool, fill roles quickly, and appeal to engineers who’ve already opted out of office life. There’s also the cost factor. Even if you’re still offering SF-level salaries, you can often make your runway go further when your team isn’t competing for real estate or spending hours commuting.
And culturally, remote gives you a head start on documentation, clarity, and asynchronous decision-making – things that matter as you scale, regardless of geography.
If you’re building heavy infrastructure or doing work that’s more experimentation than execution, that kind of clarity and autonomy can be a huge advantage.
What are the considerations when hiring remote?
That said, remote has its costs, especially for early-stage AI startups trying to iterate quickly on novel ideas.
A lot of the magic in early Machine Learning teams happens between meetings. It’s the spontaneous whiteboarding. The ten-minute hallway debug session. The moment when your research engineer hears something from product and suddenly rethinks an assumption in the model.
Those moments are harder to recreate on Zoom.
Fully remote teams can struggle with alignment —-not because people aren’t capable, but because context-sharing takes more effort. Distributed teams need to over-communicate by default, and many just don’t. Time zone lag creates friction. Collaboration starts to feel transactional. And when things break (as they always do in early-stage ML), it’s harder to respond quickly as a unit.
We’ve seen promising ML hires fail to ramp simply because they never felt part of the team. Not because they were junior — but because they were 12 hours ahead, waiting two days for feedback, and missing the nuance that happens in the room.
What are the benefits of hiring on-site?
For companies solving research-heavy problems or building brand-new infra, having everyone in the same room still matters.
On-site teams benefit from shared momentum. When the model’s not training right, you can pull someone over. When an experiment looks weird, you can walk through it on a whiteboard. When you’re hiring, candidates get to feel the pace, the energy – the real culture, not the Zoom version.
There’s also a confidence that comes from seeing things happen in front of you. For first-time founders, especially those still learning how to run ML teams, that visibility is valuable. You catch problems faster. You feel progress more tangibly. And when you need to pivot, the feedback loop is tight.
Engineers who are early in their career often grow faster in these environments, too. They soak up context. They watch how decisions get made. They sit in on conversations they might never be invited to remotely.
But “on-site” doesn’t mean what it used to
Even for companies that want people in the room, we’re not going back to the old model of five days a week in the office.
Most startups we work with are converging on a hybrid rhythm – 2-3 anchor days a week, core hours, optional Fridays. They’ve realised that flexibility still matters, even when co-location is the default. What they care about isn’t attendance for its own sake, it’s whether people can build fast, together, without needing to schedule a calendar invite for every interaction.
The best hybrid setups make room for deep work and high-bandwidth collaboration. They create overlap without enforcing rigidity. They treat in-person time as a tool, not a rule.
What Machine Learning candidates want from their next role?
In 2021, remote was a hiring superpower. In 2025, it’s just a baseline.
Most ML engineers expect some flexibility. But among senior talent, especially those who’ve worked in remote-heavy orgs, we’re seeing a quiet return to in-person curiosity. Not a rejection of remote, but a willingness to go back into the office if the problems are interesting enough and the team is worth it.
What engineers really want is clarity. If you’re a remote company, own that. If you’re hybrid with core days, say that up front. Candidates don’t want to be surprised. They want to know how work gets done, how decisions get made, and how they’ll be supported.
And the more honest you are, the more likely you are to attract the right kind of people for your model.
What this means for founders
Your stance on remote vs. on-site hiring isn’t just a logistics choice. It’s a strategic one. It shapes how you recruit, how you build, and how you grow.
If you’re early-stage and trying to hire your first few ML engineers, you might benefit from getting people in the room. If your company’s growing and needs to pull talent from outside SF, leaning remote might make more sense. If you want both – clarity becomes even more important.
Whatever you choose, be decisive. A fuzzy policy won’t help you hire. Engineers want to know what they’re walking into.
Remote and on-site both work – just not by accident.
Need help building your Machine Learning team in San Francisco?
Codexo partners with early-stage AI startups to attract and close exceptional ML talent – whether you’re building in a warehouse in SoMa or across three time zones.
If you’re hiring, we can help. Get in touch today to find out more.