Forem: Frances

AI Agent Workspace: Every Customer, No CRM Software

Frances — Tue, 28 Apr 2026 19:11:53 +0000

Every active customer has a workspace. It contains everything — their profile, lifecycle stage, onboarding history, follow-up notes, and a running log of every interaction. No CRM, no subscription, no fields I'm supposed to fill in but never do.

A customer workspace in Waxell Connect is a persistent, agent-readable environment where all context for a single customer lives: their files, their state, their history, and the playbook that tells every agent how to work with them. Unlike a CRM record, the workspace is active — agents read from it, write to it, and make decisions from it without anyone copying information into a prompt.

I used to have a CRM. It was fine. I even kept it current, for about three months, until I didn't. The problem wasn't the software — it was the workflow. Every interaction meant a context switch: finish the call, open the CRM, fill in the fields, return to work. When it was time to send a follow-up email, I'd open the CRM again, pull up the notes, paste the relevant parts into a chat with my AI, write the email, send it. Then update the CRM to say the email was sent.

That's five steps for one email. Four of them are moving information from one place to another.

The workspace-per-customer setup

One workspace per active customer. Each workspace has four things.

A profile state object — their name, company, package, timezone, use case, and any specifics about how they prefer to communicate. Not a document. A state object is a live, versioned, agent-readable data structure. When their package tier changes, I update it once. Every agent entering their workspace reads the updated version automatically on its next run.

A lifecycle stage field in the same state object — "onboarding," "active," "at-risk," "churned." When the stage changes, a scheduled task fires and creates the right follow-up sequence. Built the trigger once.

A history file — a running log of every meaningful interaction: support tickets, feature requests, things I noticed in calls. Agents append to this file. I read from it before calls. It stays current without anyone managing it.

A workspace playbook — the brief for any agent entering this space. Who this customer is, what they've asked for, what to watch for, what to avoid. Written once, read every time.

What the email workflow actually looks like

A customer sends me a support question. A scheduled task checks each customer workspace for new inbox items twice a day. When it finds one, it:

Reads the profile state object
Reads the history file for relevant prior context
Reads the playbook
Drafts a response

The draft lands in the workspace channel. I read it, edit it if I need to, send it. If I don't need to edit it — which is most of the time — it goes out as-is.

The agent already knew who this person was. I didn't paste anything.

Where Connect differs from a CRM

I'm not arguing CRMs are wrong for every situation. For a sales team tracking a pipeline across multiple reps, with quota reporting and activity logging and manager dashboards, a proper CRM earns its keep.

But I'm one person running all customer relationships myself. My reporting needs are: who is in what stage, who needs attention this week, what did I last say to each person. I have a table in a shared workspace that tracks every active customer: name, company, stage, last contact date, next action. Agents update it when they complete tasks. I review it Monday mornings.

The actual work — the emails, the follow-ups, the context behind those emails — happens in the individual customer workspaces, not in the table. The table is the summary layer. The workspaces are where the knowledge lives.

A CRM stores data for humans to retrieve. A workspace stores context for agents to act on. There's overlap, but the center of gravity is different.

What holds this together

State persistence. The agent entering Maria's workspace doesn't need me to tell it who Maria is. That's in the workspace, structured to be read, and it's the same data that was there last week. When something changes, I update the state object once. One change, everywhere it matters.

I've run this for about five months. The thing I didn't expect was how much time I'd been spending just finding context before — not doing anything with it. Before a call now: open workspace, read history file, five minutes. Before: open CRM, open notes doc, open email thread, try to piece together what the last conversation was about — twenty minutes if I was honest about it.

What you could build

One workspace per thing you track over time. The same pattern works for freelance clients, product SKUs, job candidates in a hiring process. The question worth asking: what context do I re-explain every time I work on this thing? Whatever that is belongs in a workspace state object, not in your head.

If you want to start somewhere, build one customer workspace and run it for two weeks before deciding whether to roll it out across your full list. Early access to Waxell Connect is at waxell.ai/get-access.

FAQ

How is a Waxell Connect workspace different from a CRM record?

A CRM record stores data for humans to retrieve. A Connect workspace stores context for agents to act on directly. When an agent enters a customer workspace, it reads the playbook and state objects automatically — it arrives knowing who this customer is, what's happened, and what to watch for. It doesn't wait for instructions.

Can I track customer stage and pipeline in Connect without a dedicated CRM?

Yes, if your pipeline is simple and your team is small. I use a table in Connect that shows stage, last contact, and next action for every active customer. That's enough for a one-person operation. For a sales organization that needs quota tracking, forecasting, and activity logging by rep, Connect doesn't replace Salesforce.

How do I set up a lifecycle stage trigger in Connect?

Add a lifecycle_stage field to the customer's state object. Build a scheduled task that checks whether the stage has changed and, if it has, creates the follow-up items for the new stage. First-time setup takes about an hour. After that, it runs on its own.

What goes in a customer workspace playbook?

The things you'd tell a colleague covering for you: who this customer is, what they're trying to accomplish, what's worked, what hasn't, how they prefer to communicate, what to avoid. Keep it under 500 words. Longer playbooks tend to bury the important things in the middle.

How do agents update the customer history file automatically?

Include an instruction in the workspace playbook telling agents to append a brief summary to the history file when they complete a task. Agents do this reliably when the instruction is in the playbook and the file already exists. You have to create the file first — agents won't generate it from nothing.

Do I need to be online when customer tasks run?

No. The twice-daily inbox check runs on its schedule. The lifecycle follow-up sequences fire when stages change, not when I remember to trigger them. That's the point.

Sources

Salesforce. State of CRM 2025. https://www.salesforce.com/resources/research-reports/state-of-crm/
HubSpot. CRM and Sales Statistics 2026. https://www.hubspot.com/marketing-statistics

Why Your AI Agent Forgets Everything When You Close the Tab

Frances — Wed, 22 Apr 2026 13:48:13 +0000

I spent months re-explaining myself to an AI that couldn't remember me. Every session: who I am, what I'm building, what the voice sounds like, what the customer context is, where the project stands. Paste it in, do the work, close the tab. Open a new one. Start over.

Waxell Connect solves this by giving AI agents a persistent place to work between sessions. A workspace in Connect contains everything an agent needs to pick up where it left off: files, state objects, playbooks, and task history — all structured so agents read them automatically on entry. Context doesn't disappear when the tab closes. Work accumulates across sessions. Nothing has to be re-explained.

It wasn't a model problem. It was an architecture problem — and it has an architectural solution.

The Problem Has a Name

Every AI agent operates by default in session-only memory. The model does good work within a conversation — it tracks everything said in the exchange, reasons across long contexts, builds on earlier points. But when the session ends, that context doesn't go anywhere. The next session starts fresh.

For a one-off question, fine. For work that compounds over time — a customer relationship, a content strategy, a running set of processes, a product roadmap — it creates a tax. Every session begins with re-establishment. I was doing this for months before I tracked the time: ten to fifteen minutes at the start of every AI session just getting the model oriented before doing any actual work.

The re-briefing tax is slow and inconsistent — two separate problems. What I paste in on Monday isn't exactly what I paste in on Thursday. The context drifts. The agent's understanding of my voice, my priorities, my customers is whatever I happened to include in today's prompt, not a fixed record of anything.

And the deeper problem: none of that context lives anywhere. When the session closes, it's gone. The work the agent did — the reasoning, the decisions, the output — exists only in a chat window or in whatever I managed to copy somewhere before closing the tab.

The scale of it isn't small. OutSystems' 2026 State of AI Development research found that 96% of enterprises are already running AI agents in some capacity — meaning this structural overhead is playing out across entire organizations, not just individual workflows.

A better model doesn't fix this. The capability is already there. What's missing is a persistent location for context to live between sessions.

The Connect Answer

The alternative is to stop storing context inside chat sessions — and start storing it in a workspace.

A workspace in Waxell Connect is a persistent environment where files, data, and context live between sessions. When an agent enters a workspace, it reads what's there: the playbook, which contains the brief; the state objects, which contain the current data; the files, which contain the standards, the history, the reference material. It doesn't need to be told what the workspace is for — it reads that, the same way a new hire reads a shared drive before their first meeting.

The difference is that a workspace is designed for agents, not just humans. Files are structured to be agent-readable — consistent format, clear purpose, positioned as the source of truth rather than a reference someone made once. State objects are live data objects, not static documents: agents can query a state object, update it when something changes, and build decisions from it. Scheduled tasks can read from the workspace and write output back to it without anyone being online. Channels let agents post updates, surface decisions, and hand off to humans — or to other agents — outside of a chat window that disappears.

Write the context once. Don't explain it again.

What Changes When Context Persists

The obvious change is time. I'm not spending the first chunk of every session re-establishing context. Across every workflow, every week, that adds up — and that time was the whole point of using AI to begin with.

The more important change is accuracy. When brand voice guidelines live in a workspace playbook instead of my clipboard, every agent that touches that workspace uses the same guidelines — not my best recollection of them on a Tuesday morning. When a customer profile lives in a state object instead of a preamble I paste into a chat, the agent working that account is working from the same picture I have, updated to reflect the current state of the relationship. Project status lives in a table, not in my head — so the next task picks up from exactly where the last one left off.

Context that lives in a workspace is the actual thing: maintained in one place, always current, not a reconstruction of what I happened to paste in that morning.

There's a compounding effect that takes a few weeks to feel. Update a playbook and every future session reflects it — one edit, not a dozen re-briefings. When an agent writes output back to a workspace file, the work didn't disappear — it's there, versioned, available to the next task in the chain. The workspace accumulates with every session. That's not how starting from zero works.

Where to Start

One workspace. One playbook. One piece of context you're currently re-typing every session.

Pick the workflow you repeat most. Create a workspace for it. Write a playbook that contains what an agent needs to start working immediately — the purpose, the voice, the standards, the current state. Move your most-referenced data into a state object rather than a block of text you paste in each session.

From there it scales: one workspace per customer, one per project, one per recurring workflow. Each one is an environment where context accumulates rather than resets. Each one is ready when an agent arrives.

The tab still closes. The work doesn't.

Start here: waxell.ai/get-access.

FAQ

Why does my AI agent forget what we talked about in previous conversations?
AI agents operate by default in session-only memory — context exists within a conversation but doesn't survive when it ends. Changing models doesn't fix this; it's structural. The solution is to store context in a persistent environment like a Waxell Connect workspace, where files, state objects, and playbooks live between sessions and agents read them automatically on entry.

What is a workspace in Waxell Connect?
A workspace is a persistent environment where files, data, and context live between sessions. When an agent enters a workspace, it reads the context that's there — the brief, the standards, the current data — without anyone re-explaining the setup. Work accumulates across sessions rather than starting fresh each time.

What is a state object, and how is it different from a document?
A state object is a live, versioned data object that agents can read, write to, and act on. Unlike a document — static text that a human reads — a state object is structured so agents can query its current value, update it when something changes, and use it to drive decisions. A customer's lifecycle stage as a state object means every agent touching that workspace sees the same current picture.

What is a playbook in Waxell Connect?
A playbook is a markdown file in a workspace that agents read automatically when they enter. It contains whatever context the workspace's work requires: purpose, voice, process, standards, relevant links. The practical difference from a prompt: a prompt lives in your head and you re-type it each session; a playbook lives in Connect and agents find it. Update it once, and every future session uses the updated version.

How do I make my AI agent remember context between sessions?
Store your context in a workspace rather than in chat history or a copy-paste workflow. Voice guidelines, customer data, project state, process standards — these belong in workspace files and state objects that agents read automatically. The workspace is the persistent layer that survives when sessions end.

Can AI agents do work without me being online?
Yes — scheduled tasks in Waxell Connect run on a set schedule without anyone present. They enter a workspace, read the current context, do work, and write output back — so the next task or session picks up from the current state rather than starting from scratch. This is what makes multi-step automated workflows possible: each step reads from and writes to the workspace, which persists across all of them.

Sources

McKinsey & Company. The Economic Potential of Generative AI: The Next Productivity Frontier. June 2023. https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier
OutSystems. State of AI Development 2026: The Move to Agentic AI. 2026. https://www.outsystems.com/1/state-ai-development/