When I checked my AI usage dashboard, two numbers made me stop:
145 million tokens. 8,377 requests.
I don’t code.
Vijay, our founder and main developer at 123sudo, handles that side of the work. These were my conversations.
In September, I wrote about those figures on Medium. My 9xchat dashboard showed an all-time cost of about $637. That wasn’t my complete AI spending for the year. Including other AI tools and services, I estimated I was approaching ₹1 lakh.
Those are different totals: one was a dashboard figure, the other my broader spending estimate.
But both raised the same question:
What was I actually getting from all this usage?
It wasn’t one enormous task
I’m a co-founder working on product experience, creative direction, and growth.
My AI usage comes from the ordinary work of building and growing products:
- Working through product ideas.
- Researching an audience.
- Drafting launch content.
- Finding clearer language for a confusing feature.
- Comparing possible approaches.
- Asking another model to challenge an idea.
Sometimes I need a draft. Sometimes I need a second opinion. Sometimes I haven’t understood the problem well enough to ask a useful question yet.
The total wasn’t one dramatic project. It was thousands of small interactions.
That’s what surprised me: you don’t need to be writing code for AI usage to become a meaningful expense.
A big number doesn’t automatically mean waste
Some conversations help me make a decision. Others help turn an unclear idea into something we can work on.
That can be valuable.
I don’t want to reduce AI usage simply to make a dashboard number smaller. A useful conversation can be worth more than several cheap but unhelpful ones.
The question I’m learning to ask is:
Did this interaction move the work forward?
A finished brief, a clearer decision, or a usable draft is different from another page of options I won’t act on.
Starting over is the part that frustrates me
I expect to explain a new task.
What gets tiring is rebuilding the background around it:
What is 123sudo? Which product are we discussing? Who is it for? What have we already decided? Which claims should we avoid?
Then I switch models for another perspective and repeat the setup.
There are two costs here:
- The usage cost, depending on the model, billing method, and context processing.
- My time and attention, spent reconstructing information instead of doing the next piece of work.
I don’t have a measured savings figure for fixing that. But it’s a real source of friction in my workflow.
Remembering context isn’t the same as making it free
This is an important distinction.
Persistent memory can reduce the amount of background I manually repeat. It does not mean remembered information becomes free for a model to process.
Relevant context may still be included in a request. Its cost depends on the implementation, model pricing, caching, and other factors.
So I think of memory as a continuity tool first, not a guaranteed token-saving mechanism.
The useful question is:
Are we bringing the right context into the task, or just carrying more text?
Three questions I’m using to judge my AI work
Seeing those usage figures made me think more carefully about what a productive session looks like.
1. What am I trying to finish?
“Help me think about marketing” can become an endless conversation.
“Help me draft a product brief for this audience, using these confirmed facts” gives the work a clearer destination.
Exploration still has a place. I just want to notice when I’m exploring and when I’m supposed to be finishing.
2. What context actually matters?
A model doesn’t need every discussion we’ve ever had.
It needs the information relevant to this task: the audience, goal, constraints, and current decisions.
Old plans can be just as distracting as missing information.
3. Is another response helping, or delaying the decision?
This is the uncomfortable one.
Sometimes asking for more options is useful. Sometimes I’m asking because choosing an option feels harder than generating another ten.
AI can help me think. It can also give me a very convincing way to stay busy.
Why this matters to what we’re building
Disclosure: I’m a co-founder of 123sudo, the team behind 9xchat. I use it daily, so this is not an independent product review.
9xchat brings multiple models, persistent memory across models, and transparent credits into one AI workspace.
Those capabilities matter to me because my work doesn’t stay in separate boxes. Research turns into a product decision. That decision turns into a brief. The brief turns into content.
I want the work to continue without rebuilding the background at every step.
But having a workspace doesn’t remove my responsibility to give clear instructions, review outputs, and decide what deserves another iteration.
A tool can support a better workflow. It can’t make every conversation worthwhile.
The lesson wasn’t “stop using AI”
It was stop treating usage as proof of progress.
145 million tokens tells me how much activity happened. It doesn’t tell me how many good decisions I made, what shipped, or which conversations were worth their cost.
That’s the distinction I want to get better at.
I don’t code. AI has still become a substantial part of my working life.
Now I’m trying to judge it less by how much I use it, and more by what the work becomes.
AI helped structure and edit this article. The usage figures and personal observations are mine.
What tells you an AI session was worth it: a finished task, a better decision, time saved, or something else?
Top comments (0)