Why AI Can't Feel the Room: Practical Operations with Carlos Almansa

"The AI cannot feel the space. It can't feel the dynamics or the vibe. But it can free up time for you to talk to your members, to have a coffee with them, to understand and to read people."
— Carlos Almansa Ballesteros

Episode Summary

Most conversations about AI in coworking are either evangelical or dismissive. Carlos Almansa Ballesteros, co-founder of Nexudus and author of the Coworkings AI newsletter, isn't interested in either.

He's been building software for this industry for 13 years. That gives him a specific kind of authority: not hype, not theory, but a long, sober look at what operators actually do and where they keep getting stuck.

His answer to AI is the same one he'd give about any tool. Start with the thing you're already doing. The Wi-Fi question you answer forty times a week?

That's where you start. Not with a grand automation strategy — with the inbox that's eating your afternoon.

Then there's the line you don't cross. The first 60–90 days of a new membership. The kitchen conversation. The introduction on day one. Those stay human. Full stop. Because when someone cancels after six weeks, it's rarely the software that failed them.

No frameworks. No magic. Just what actually works.

Timeline Highlights

00:02 — Bernie sets the episode up: practical AI, not rocket science

01:16 — Carlos introduces himself: Nexudus co-founder, Coworkings AI newsletter author

03:02 — Carlos's first coworking space, and why he's always joined one when moving to a new city

06:41 — The first 60–90 days: how a community manager makes or breaks early membership

09:09 — The London Coworking Assembly AI survey: most people use it for social media captions and don't go further

10:15 — Why the Coworkings AI newsletter exists: cutting through noise to find usable signals for operators

12:54 — The solo operator and AI: actually easier to start when you know all your own processes

15:03 — Practical use case one: automating repetitive help desk replies (Wi-Fi, printing, FAQs)

16:21 — Practical use case two: surfacing data patterns you can't see manually

17:53 — The soul-of-the-space question: automation versus presence

19:36 — Sentiment analysis: feeding community messages into AI to understand the pulse of a space

20:52 — Context is everything: how to give an AI model what it needs to work properly

23:56 — What goes wrong: people who automate everything at once and erode trust

25:21 — The human-in-the-loop rule: never hand your reputation to an unmonitored system

25:56 — Transparency: be honest about AI, always offer a route to a real person

28:42 — Can you automate community? Carlos on what AI can and cannot do in a 300-person space

30:05 — The kitchen conversation: serendipity as the baseline unit of community

31:09 — What community actually is, from random coffee chats to self-organising hackathons

33:47 — Mobile work and the future floor plan: what happens when nobody needs a desk?

Lesson 1: Only Take AI to Something You're Already Doing

The most useful thing Carlos said in the whole conversation is also the least glamorous.

Before you touch any tool, review your processes. Do you actually know them? When you're a solo operator running a space from first enquiry to member induction, you probably do. That's an advantage.

The mistake Carlos sees repeatedly is that people want to automate first and understand later. You end up with ten tools stacked on top of a process that was working fine.

Start somewhere specific. The help desk is a good example.

The same questions come in every week

How do I connect to the Wi-Fi?
How do I print?
Where's the code for the meeting room?

That's exactly the kind of repetitive, low-stakes task where AI earns its place. Automate the reply. Free yourself for the conversation that actually matters.

Carlos, on the starting point:
"The best way to start with AI is to start with something that you are already doing, not trying to implement a new process or a new tool into something that you are not familiar with."

Here is how I look at it: don't ask AI to do something you don't already know how to do. I wouldn't go to AI and say, "Make me a hit record." I go to AI and say, "How do I make a podcast intro?"

Lesson 2: The First 60–90 Days Are on You, Not the Software

When a new member joins, the community manager is doing something no platform can replicate.

They're introducing names. They're reading the room. They're noticing who eats lunch alone and who lights up when someone mentions their industry. Carlos joined coworking spaces in South Spain, Madrid, and London. Every time, the spaces that made it work did the same thing: they put a person in the room who paid attention.

Shared lunches in Madrid led to basketball at weekends. Hot desking in London broke the ice faster than any directory ever could.

The first 60–90 days determine whether someone stays. Carlos is clear on this. And he's equally clear that no amount of automated onboarding email replaces what happens when you walk someone through the kitchen on day one and explain — face to face — how to clean up after yourself.

That small thing sets the tone. It says: " This is a shared space. We're in it together."

AI can surface patterns. It can flag a member who hasn't engaged in three weeks. It cannot do the introduction.

Lesson 3: Your Community's Messages Are Data — Use Them

Members tell you how they're feeling all the time. The problem is they're doing it across email, WhatsApp, Slack, and every other channel simultaneously.

Some of it is friction: "the Wi-Fi's broken again." Some of it is gold: "that event last week genuinely changed something for me." Most of it sits in inboxes, unanalysed, until the member quietly cancels their membership.

Carlos's newsletter recently covered this directly. You can feed your community's communications into an AI, build a sentiment analysis across those messages, and surface patterns that would take weeks to spot manually. Who's frustrated? Who's energised? What topics keep coming up?

That's not automating the community. That's clearing the brush so you can see what's actually there.

Carlos:
"When you have a community of 200 or 300 people in your space, it's not easy to get to know everyone, to know where they are, how they're feeling. AI can help to surface issues and connect people.the But it cannot automate those relationships."

The distinction matters. AI as an insight layer, not as a relationship substitute.

Lesson 4: Never Let AI Go Solo — Your Reputation Is on the Line

Carlos is consistent on this throughout the episode. New technology, new risks. The worst thing you can do is automate a customer-facing process and walk away from it.

If AI is replying to enquiries through your website, you need to be watching those replies for months. The system might not be calibrated yet. One bad reply to a prospective member and you've made a first impression you can't undo.

The related point is transparency. When someone lands on a chat widget, don't dress the AI up with a human name — "Hi, I'm Brad." Carlos didn't say it exactly like that, but Bernie put it plainly: that's obvious, and it immediately damages they can. Be honest. Tell people they can get fast answers here and offer a clear path to a real person when they need one.

Some people just want the answer. Some need the conversation. Build for both.

Carlos:
"You're putting your reputation as a space, as an operator, on the line. You are delegating that into a system that might not be streamlined at the first place. So you need to refine, and you need a human in the loop, at least for a while."

Lesson 5: Give the AI Enough Context to Actually Be Useful

There's a reason most operators get frustrating results when they first try AI tools. They don't give the model enough to work with.

Carlos's analogy is the right one: giving context to an AI is like training a new team member on the front desk. You wouldn't put someone on reception without telling them what the space offers, what questions come up constantly, how you like to communicate with members, and ,what the common issues are with the systems. You'd walk them through it. AI needs the same treatment.

For community sentiment, that means feeding in your actual communications — the tickets, the WhatsApp messages, the emails. For website enquiries, it means giving the model your pricing, your offering, your tone of voice. The more specific the context, the more useful the output.

This is also why Carlos says solo operators may actually have an advantage. When one person runs everything, the processes live in one head. There's no scattered handover across a team, no misalignment. You know what you do and how you do it. Translating that into context for an AI is straightforward once you decide to sit down and do it.

Carlos:
"The context the AI needs — if we were to narrow it down — understanding the sentiment of the community based on the messages they sent, you need to feed the messages, all the communications you have, all the tickets the community raised with you. It's similar to training someone who joins your team at the front desk. Start with one use case. Write down the context for that specific thing. Test it. Refine it. Then move on."

Links & Resources

One More Thing

I've been in this industry long enough to have watched every wave of technology get positioned as the thing that would finally fix the community.

It doesn't. It can't.

Carlos builds software. Thirteen years of it. And the line I keep coming back to from this conversation is one of his most straightforward: the AI cannot feel the space. It's not a limitation he's apologising for. It's a description of reality.

What AI is good at is the grind. The same Wi-Fi question forty times a week. The member's message gets buried under twelve other messages on a Tuesday afternoon. The data patterns that would take you three months to see if you were looking at them manually. Handle that part well, and you get something back: time. Time to be on the floor. Time to notice who's struggling, who's about to leave, who's one good conversation away from their next thing.

That is the actual job.

The spaces I've seen get community right didn't have a community strategy. They had someone paying attention. They had a kitchen that people actually used. They had a culture where it was normal to ask a stranger what they were working on.

AI doesn't build that. But it can stop you from eating the time you'd otherwise spend on it.

That's the deal. Use it right.