Over the past few weeks, we have shared guides and use cases showing how the coworking industry can extract real value from AI. Some carried a technical component that could slow down implementation, but the goal was worth it: showing the new value scenarios AI opens up for operators.

This time, we want to remove that friction. In this edition, we dive deep into the two latest AI features in Nexudus, where the implementation and configuration layer is minimal, so operators get immediate value from AI. The platform extracts the knowledge already living in an operator's account (FAQs, plans, resources, bookings, contracts, and more) and turns it into a context layer, a kind of space brain. From that layer, two features deliver the value:

Let's dive in and explore how they work.

The AI Assistant is a conversational front-desk agent. It answers questions, helps members manage their bookings, and guides prospects toward buying a day pass, signing up for a membership, booking a meeting room, or scheduling a private office tour.

The assistant works across four channels: Chat, Email, WhatsApp, and Voice. All four share the same capabilities, the same data, and the same visibility rules. The differences lie in session handling and response formatting.

Identification depends on the channel. In the portal, the AI recognizes members automatically through their login. On email, it matches the sender's address to a customer record. On WhatsApp and voice, it matches the sender's phone number. When the AI performs an action that changes data (creating or canceling a booking) it asks for confirmation within a channel-specific window: 10 minutes on chat, 1 hour on WhatsApp, 4 hours on email.

The standalone AI Chat widget embeds the same assistant on any external website (a company homepage, a landing page) with a single script tag. Conversations started there appear in the AI Conversations list alongside portal conversations, giving the operator a unified view of every AI interaction, wherever it started.

When a customer or prospect sends a message, through the chat widget, voice, email, or WhatsApp, the AI identifies the right tool for the request (searching FAQ articles, looking up available meeting rooms, checking membership plans), calls that tool to fetch live data from the operator's Nexudus account, and writes a reply based solely on what the tool returned. It maintains context across turns, so a follow-up like "book the second one for tomorrow" works correctly.

Every answer is grounded exclusively in the information available in Nexudus, which lowers the risk of invented or inaccurate answers.

The AI only shows what the operator allows. The visibility settings add four AI-specific fields to day pass products, membership plans, and private offices:

In the Nexudus admin panel, a dedicated Settings > AI > Items page shows every product, plan, and floor plan item in one table with inline toggles, and an "AI Ready" badge marks enabled items throughout the admin panel. Beyond these fields, the AI follows exactly the same access rules the operator already manages in Nexudus (the FAQ access setting, the visibility flags on each resource). There is no separate AI permission layer to maintain.

All capabilities work on all four channels:

Conversion tracking is what turns the AI Assistant into a measurable sales channel, not just a support tool. Every conversation carries a unique session ID, and that ID is embedded in every call-to-action link the AI generates. When someone completes the action, Nexudus records the link between the outcome and the conversation that produced it.

Five types of conversions are available: Tour arranged, New customer created, Membership started, Booking made, and Day pass purchased. A single conversation can produce multiple badges — a new prospect who signs up generates both a new customer and a new contract. Conversations with no badges are still useful: they may represent members asking questions, or prospects who browsed but haven't yet converted. Reviewing unconverted conversations can surface friction points in the sales process.

In the Nexudus Admin Panel, under Operations > AI Conversations, the operator sees every conversation from every channel in one list (the opening message, the channel, the conversion badges, the identified user). Clicking a row opens the full transcript. When a prospect books a tour through the AI, the complete conversation is also embedded on their Visitor record, so the team knows exactly what the prospect asked about before the visit.

The system works from the operator's own data, respects the access settings already configured in Nexudus (members-only content, hidden pricing, archived items), and logs outcomes so the operator can see which conversations actually drove revenue.

Now let's turn to Proactive AI Agents.

The AI Assistant is reactive: it waits for visitors or members to ask. Proactive AI Agents work the other way around. They are an operator-facing intelligence layer in the Dashboard that monitors patterns in the space (an overdue invoice, a contract approaching expiry, a member who hasn't checked in for two weeks) and automatically proposes the right action for review.

In previous Coworkings AI editions, we explored what a skill is. A skill is a structured document that teaches an agent how to handle a specific type of task. It captures a repeatable way of working: the steps someone follows, the judgment they apply, the decisions that have been made over time. It can include step-by-step instructions, success and failure examples, edge cases, hard constraints, escalation criteria, and reference outputs from the team's best work.

Proactive Agents embed that same skill logic: each agent is a step-by-step process designed to deliver a specific outcome.

Each agent monitors for the patterns defined in its criteria, evaluates any match against a confidence score, and proposes an action in the AI Inbox for the operator's review, before executing it once approved. The message goes out by email, WhatsApp, voice call, or stays as an internal note. If auto-execute is enabled, high-confidence actions are sent without manual review.

Agents are organized into seven categories, each reflecting an area of the business they monitor:

The Proactive Agents system sits alongside the conversational AI Assistant, and the two are connected bidirectionally. The assistant handles real-time customer questions in the Members Portal; the agents use that same conversation data, along with booking, invoice, and contract data, to detect patterns. When an agent identifies an opportunity, it proposes an action through the AI Inbox for the operator's review. Once an approved email, WhatsApp, or voice action goes out and the customer replies, that reply starts a two-way conversation handled by the same conversational AI, visible in AI Conversations and linked to the original proposed action so the full context stays in one place.

Once an operator has identified which proactive agents are right for the space, the next step is configuring each one to match its operational needs. The configuration panel controls when an agent acts, how often, through which channel, and what tone its messages use.

In the Nexudus admin panel, configuration lives under Operations > AI Agents, where each agent appears as a card with its status, category, channel, cooldown, and rate limit. Configuring an agent comes down to a handful of controls:

The Agent Types Reference documents every agent (what it monitors, when it triggers, and its recommended configuration settings). Four are live today:

The remaining ten: FAQ Gap Detection, Lead Follow-up, Onboarding Nudges, Tour Reminders, At-Risk Member Detection, and Upsell Opportunities, will be released soon.

This agent answers a question every operator has asked: is this week's flood of tickets normal, or is something broken?

The agent detects statistically significant spikes in support issues within specific categories compared to a rolling average baseline. When a spike is detected, it creates an internal alert for operators, not an outbound message to customers, to flag a potential systemic issue requiring attention.

The agent automatically classifies support tickets into nine categories using AI analysis (Printing, WiFi, Access, Billing, Noise, HVAC, Cleaning, Booking, or Other). It then compares each category's current weekly ticket volume against a rolling average baseline (four weeks by default). When a category exceeds the baseline by the configured multiplier (2x by default) it creates an internal alert: "Printing issues increased 300% this week (12 tickets vs. average 3). Possible printer problem."

It has the same anatomy as any skill: a problem it addresses (systemic issues hiding in scattered tickets), a workflow it encodes (classify, baseline, compare), a threshold for success (a spike beyond the multiplier with sufficient baseline volume), and a data source it draws from (the space's Help Desk). It requires a minimum baseline of tickets before alerting, so low-volume categories don't generate noise. It sends one alert per category per week, and it never contacts customers directly, the output is an internal note for the team. That makes it one of the few agents an operator can safely run on auto-execute from day one.

The AI Inbox is the central workspace where proposed actions land. It works like a drafts folder: the AI prepares each message, and nothing reaches a customer until the operator reviews it. It is worth distinguishing it from AI Conversations:

The Inbox gives the operator full control over every message: approve as-is, edit for personalization, reject with feedback, or snooze for later. Over time, auto-execute can be enabled for trusted agents so high-confidence actions send automatically without manual review.

Once a proactive agent detects a pattern worth acting on, a proposed action appears in the AI Inbox. The left panel gives the operator everything needed to decide whether the action makes sense (customer information, why it was proposed, the criteria, and the confidence score). For example: "Invoice #INV-2024-0847 is 5 days overdue. Customer has a strong payment history but missed this payment, which is unusual." That context, together with the customer's recent activity, payment history, and open support tickets, prevents mistakes like sending a payment reminder to someone who just filed a dispute.

The right panel is the message draft itself, fully editable. The operator has four options: approve and send, edit and approve, reject, or snooze for later.

Operator Takeover lets the operator change how a proposed AI action is delivered to the customer. When reviewing an action in the AI Inbox, the operator is not limited to the AI's suggested channel and can choose:

All messages, regardless of channel, are logged in the conversation timeline in AI Conversations, so the full case history lives in one place. From there the operator can step in at any point: the conversation history, the trigger that started it, and any resulting Help Desk ticket are all connected, so nothing gets picked up cold.

Both features depend on the same foundation: AI only works reliably once the context layer behind it is in place — here, Nexudus's own data (knowledge base articles, membership plans, bookings, and signed contracts). From there, the two features pull in different directions:

The AI Assistant helps members have a better experience in the space, while Proactive AI Agents help operators streamline and optimize daily operations.

Both are already available, free of charge while in beta. If you are interested in trying them, you can write to us here.

That's it for today. See you in two weeks.