Switching models is easy. Keeping your decisions, project context, and next steps connected is the harder problem.
You ask one AI to research an idea, another to challenge it, and a third to help you write.
Then you become the courier.
You copy the brief. Explain the decisions. Remind each tool what you already ruled out. Somewhere along the way, you lose the version you liked and start searching through old conversations.
The answers might be useful. The process still needs work.
I know this frustration because it’s part of why we started building 9xchat at the end of 2024.
Disclosure: I’m a co-founder of 123sudo, the team behind 9xchat. This is both a workflow problem I care about and a product I help build.
Back then, I could feel my work becoming tied to one AI app. My chats were there. My project context was there. The background I had spent time explaining was there.
Trying another model was possible. Taking that accumulated understanding with me was harder.
That made me question what I actually needed from AI. Was it another place to ask questions, or a better system for continuing the work?
The work between the answers
A quick question doesn’t need much continuity.
A project does.
When you’re building something over days or weeks, each conversation sits inside a larger story. There’s an audience you chose, an approach you rejected, a deadline you’re working toward, and a reason the latest draft looks the way it does.
An individual answer can be excellent and still miss that story.
You ask for a launch plan, and it recommends a channel you’ve already ruled out. You ask for a revision, and it changes something you deliberately kept. You start a new conversation and spend the opening messages rebuilding the brief.
None of these moments looks dramatic on its own. Together, they make working with AI feel strangely repetitive.
You’re not just doing the project. You’re maintaining the explanation of the project.
That is the work between the answers.
A better model is not automatically a better workflow
Model capability matters. Different tasks benefit from different strengths, and choosing the right model can make a real difference.
But model quality and workflow quality are not the same thing.
A more capable model still needs relevant information. If it doesn’t have your current constraints or the reasons behind your decisions, you may get a polished answer that takes the project in the wrong direction.
The question isn’t only:
Which model should I use?
It’s also:
What does this model need to know so we can continue instead of starting again?
That second question gets less attention, but it has become central to how I work.
I don’t want every conversation to begin with a longer prompt. I want a reliable way to carry forward what matters, inspect it, and change it when the project changes.
Before adding another tool, trace one task
If your AI setup feels fragmented, start by looking at one real task.
Choose something you recently completed: a proposal, an article, a research brief, a feature plan. Follow it from the first question to the finished result.
Where did you repeat yourself?
Look for the background information you explained more than once.
Was it the audience? The tone? A technical constraint? A decision you thought was already settled?
Repeated explanations show you which context needs a more deliberate home.
What did you move between tools?
Not every copy and paste is a problem. Sometimes a specialist tool is worth the handoff.
But notice when you’re moving information simply because the next tool has no other way to access it.
That’s coordination work. It deserves to be counted as part of the cost of your setup.
Which decisions disappeared?
Finding an old answer isn’t always enough. You also need to know whether it still applies.
What did you choose? Why? What changed afterward?
A saved transcript contains the conversation. It doesn’t necessarily give you a clear record of the current decision.
What did each tool contribute?
Some tools earn their place through a capability you genuinely need.
Others remain in your workflow because you’ve become used to working around them.
The point isn’t to force everything into one app. It’s to make sure the handoffs are helping you, rather than quietly becoming another job.
**
Keep the useful context, not everything**
My goal isn’t to make an AI remember every sentence I’ve ever written.
It’s to make the right information available when it matters.
For an ongoing project, I’d start with four things.
The brief: What are we making, who is it for, and what should it accomplish?
The constraints: What limits the work? Budget, tone, technical requirements, deadlines, or approaches we’ve ruled out.
The decisions: What have we chosen, and why?
The next action: Where did we stop, and what needs to happen next?
Here’s a simple example:
Project: An article for independent builders using multiple AI tools.
Goal: Help readers identify where their workflow loses context.
-Constraints: Practical rather than speculative. No invented productivity claims.
- Decision: Lead with the reader’s frustration, not the product’s features. Next action: Review the draft for unsupported claims and unnecessary promotion.
That’s a small amount of information, but it gives the next conversation somewhere useful to begin.
You can keep a project record like this in a document or a note. You don’t have to change tools to start working more deliberately.
The system matters before the software does.
Why we started building 9xchat
At the end of 2024, the problem felt clear to me: I was investing time in conversations, but too much of their value felt tied to the app where they happened.
I didn’t want choosing a different model to mean rebuilding my working context.
That frustration helped push us to build 9xchat.
Since then, it has evolved into an agentic workspace with multiple models, memory, notes, tools, Skills, and MCP connections.
But the most meaningful change for me isn’t the growing feature list.
It’s how much I now build with it.
I use 9xchat every day, and I find it hard to imagine my working routine without it. Not because it does everything for me, but because it has become part of how I develop ideas and move work forward.
That experience has made something very clear to me:
The model matters. The system you build around it matters, too.
**
What the system adds**
In 9xchat, I can switch models within a conversation rather than treating each model as a completely separate working environment.
Memory gives relevant information a way to persist beyond a single exchange. Notes provide a place for material I want to keep deliberately. Agents and reusable prompts help define recurring ways of working.
These are different jobs. A note isn’t the same as a memory, and neither replaces checking the current brief.
The distinction becomes even more important as the workspace gains more capabilities.
The public documentation separates three building blocks:
Tools let the assistant perform specific actions, such as searching the web or working with notes.
Skills provide instructions for how it should approach a task or structure an answer.
MCP connections make external tool collections available to the assistant.
Tools, Skills, and MCP use in chat are part of Pro.
To me, their value isn’t simply that the AI can do more things. It’s that they can support a more coherent way of working: relevant context, an appropriate method, and access to the tools needed for the task.
That still requires judgment. A connected tool isn’t automatically the right tool. Saved information isn’t automatically current. A successful action isn’t proof that the entire task is correct.
Being able to inspect and question the work remains essential.
**
Using it is part of building it**
At 123sudo, there’s always something to build or improve.
Using 9xchat daily keeps showing me where the next improvement should be.
Where did I have to repeat myself? What took too many steps? What did I expect the workspace to understand that it didn’t?
Those moments are useful feedback precisely because they interrupt real work.
I’m not only looking at the product from the outside. I’m inside it, trying to get something done, noticing what helps and what gets in the way.
That doesn’t make my experience representative of everyone’s. We now have more than 1,000 users, bringing different projects, habits, and expectations. Their feedback matters alongside our own.
People are also paying for access through monthly and lifetime licenses. That adds responsibility. A workspace has to earn its place through use, not just through a convincing demo.
For clarity, Pro access and AI usage are separate: the current pricing offers monthly, yearly, and lifetime Pro options, while 9xchat AI credits are purchased separately.
**
The ambition is bigger than the current product**
One day, I’d love for 9xchat to help manage and coordinate far more of the systems we use at 123sudo.
But that is a direction we want to explore, not a claim that it can run everything today.
Broader capabilities need careful development: clear permissions, dependable behavior, ways to verify actions, and boundaries around what should require a person’s approval.
Skills and MCP support are already here. What we build on top of them still needs to evolve through testing, feedback, and real use.
I’m excited about that future. I’m also interested in the smaller improvements that make tomorrow’s work less frustrating than today’s.
Sometimes that means another capability.
Sometimes it means making an existing workflow clearer, more reliable, or easier to control.
**
The milestone I’m proudest of**
9xchat is the first digital app I’ve helped build that I use every single day.
Not just to test it.
To do my own work.
There’s something meaningful about helping build a tool and then repeatedly choosing it when you have something real to finish.
I still see the rough edges. I still have ideas for what it should do better. In a way, that’s the point: using it keeps me close to the problem we started with.
I wanted my work with AI to feel less scattered.
I wanted to spend less effort carrying context and more effort building on it.
That remains the goal.
The next leap might be a more connected way of working
A more capable model can open up new possibilities. I’m not dismissing that.
But I no longer expect a model upgrade alone to fix a fragmented working process.
If your brief is scattered, your decisions are buried, and every new conversation requires another introduction, the system around the model deserves attention too.
Start with one project. Keep its brief current. Record the decisions that matter. Make the next action clear.
Then judge your setup by a practical question:
When you return to the work, can you continue, or do you have to explain it all again?
Try keeping one project’s context connected in 9xchat.