These articles come from lessons learned while building Eterna Clarity and the operating system I use to run it.
If I listed the software and AI systems I use to run Eterna today without explaining that I am a solo founder, it would sound more like the operating stack of a small technology company.
ChatGPT has become my daily driver. Claude gets a lot of coding work and writing work, with Eterna's own Studio standards around it. Grok is heavily used for Imagine, while Grok Bot handles longer-running work around areas like networking, sales and Eterna's free resources. Muse has become useful for long-running mechanical work, including the current EternaAI training programme.
Google's AI tools are particularly useful when I want another system to absorb and review very large documents or long context. Qwen has become one of the systems I like for visual review and as another adversarial perspective. Copilot gets used in a similar way when I want another independent pass. I am using all of them regularly now.
Then there is everything that does not look like an AI provider. GitHub and Cloudflare sit behind a lot of the website and application work. Supabase carries production product databases and related infrastructure. Stripe handles payments. Figma and Canva are part of the design stack. ElevenLabs is now part of Studio production. FFmpeg and Kdenlive do real media work locally. Python appears everywhere because sometimes the fastest answer is simply to write the exact program needed for a task.
There are more tools around the edges, but the important thing is that this no longer feels like a collection of subscriptions. It feels like a company operating environment.
I stopped expecting one AI to be the company
For a while, there was a natural tendency to think about AI capability in terms of which model was best. Which provider should be the main one? Which model is smartest? Which one codes better? Which one has the biggest context window? Which one should I build around?
I care much less about finding one answer to those questions now. Different providers are genuinely better fits for different work. Even when two of them are technically capable of doing the same task, the experience can be different enough that I develop preferences. One may suit the way I want to code. Another may be better for a long document review. Another may have a creative tool I cannot replace. Another may be useful because I can leave it working on a long-running mechanical task without tying up the surface I want for something else.
That became much more obvious once I started using several providers constantly rather than occasionally testing them. The more useful question is no longer which AI should run the company. It is what role each system has earned.
That also lowers the pressure on every tool. ChatGPT does not have to be my image generator, autonomous networking worker, local training system, database, payment processor, video editor and source repository. Claude does not need to own company state just because I like using it for code. Grok does not need to become the operating system because Imagine is useful. Each system can remain good at the thing I actually want from it.
The boring software is still doing a huge amount of the work
It is easy to describe this as an AI stack because the AI providers are the most visible part. That would miss a large part of where the leverage actually comes from.
A language model can help me reason through a website change, but GitHub still owns the source and Cloudflare still has a real deployment job. The production application still needs a database. Payments still need an actual payment provider. Video still has frames, codecs, audio tracks, timing and exports that are often easier to manipulate with ordinary media software than another prompt.
The same is true inside Eterna. If something is exact, I increasingly want exact software handling it. If Eterna already knows a rule, I do not need a frontier model rediscovering the rule every time. If a workflow can be represented deterministically, it can become code. If a piece of information has an authoritative owner, the AI can retrieve it instead of trying to remember it.
That is why Python, FFmpeg and all the other ordinary software around the models matter so much to me. They turn model capability into repeatable operations. There is a huge difference between having an AI explain how to do something and having a working system that now knows how to do it.
The part that makes the stack compound is underneath it
The most important part of this stack for Eterna is not any individual provider. It is the system connecting the work.
EternaAI, the Eterna Engine, the Workspace and the MCP connections underneath them increasingly provide the operating layer between me and all of these different capabilities. That layer carries things the providers should not have to own. Current company state has durable owners. Jobs and Work Packages survive individual conversations. Systems know which operations exist. The Engine can resolve where work belongs. Accepted standards and playbooks can be retrieved again. A provider can enter Eterna, get the relevant current state, do useful work and leave without taking the company with it.
That architecture has been covered in other Eterna Articles because it solves provider dependence and company-state problems. The newer consequence I am noticing is that the tools are starting to compound each other.
A difficult task may begin with frontier reasoning. Once the task is understood, some of it becomes software. The accepted workflow becomes a playbook. A failure becomes a test. A recurring research pattern becomes a reusable process. A provider discovers a better way to represent something, and that representation is available to the next provider too.
The next task therefore does not always start from the same place. If I paid for seven AI providers and every new conversation began from zero, I would mostly have seven places to ask questions. That is not what I want. I want the company to get better at using all seven.
Multiple reviewers have changed the quality bar too
One benefit I did not appreciate enough at first is how cheap independent review has become.
If I am uncertain about a visual, I can have another multimodal system inspect it. If a long technical plan feels convincing, another provider can challenge it. If one model writes something, a different one can review it against the real Studio standards instead of simply asking the original model whether its own work is good. I use Qwen and Copilot this way often. Google is useful when there is a large amount of material to inspect. Other providers get pulled in depending on the work.
None of those reviews automatically becomes correct because another AI said it. I have learned that lesson too many times already. What changes is the cost of disagreement.
Historically, getting another capable person to deeply review a technical design, visual asset, long document, product workflow or piece of code was expensive because another person's time is expensive. That is still true when genuine specialist human expertise is required. But there is now a large layer of review where I can cheaply ask another capable system to look for what the first one missed.
That does not replace judgment. It gives judgment more evidence. For a solo founder, that matters because one of the obvious weaknesses of working alone is that there is nobody sitting beside you naturally challenging your assumptions. I can create some of that pressure now without pretending an AI reviewer is a substitute for a real specialist when one is needed.
The monthly bill is the part I still find difficult to believe
The recurring software cost behind the current Eterna operating stack is only a little over CAD $100 per month.
That number needs some context. It is not the total cost of operating a company. It does not count my time, the computer sitting on my desk, taxes, transaction fees or every variable business expense. It would also change if Eterna's usage or commercial scale changed substantially.
But as the recurring software cost for the stack I am actually using to build and operate the company, it is still extraordinary to me.
For that money I have access to several frontier AI systems, coding capability, research and long-context review, multimodal analysis, image and video generation, autonomous workers, local AI, source control, production hosting, databases, payments, design systems, media production software, speech generation and the connectors that let many of these systems interact with Eterna.
Then Eterna's own software sits around those services and makes them specific to the company. There are workflows for recurring work, deterministic operations, playbooks that preserve lessons, quality standards, provider roles and systems for research, networking, sales, creative production, products and operating work. There is also an increasingly capable local intelligence layer being trained around the work that should not always require frontier AI.
I am one person, which is still the part that feels slightly absurd when I stop and look at it.
I am not claiming I have replaced an enterprise workforce. There are countless things a real team of experienced specialists would know or do better than I can. Human capacity, domain expertise, relationships, taste and accountability do not disappear because software got cheaper.
What has changed is the amount of sophisticated work one person can realistically attempt before headcount becomes the limiting factor. I can build production software, run databases, make and edit video, develop a website, conduct substantial research, build sales and networking systems, produce design work, test systems adversarially, train a local model and maintain operating state across all of those areas.
Doing those things well still requires judgment and a lot of work. The remarkable part is that access to the underlying capabilities is no longer the expensive part.
Buying more tools is not the lesson
There is an obvious bad conclusion someone could take from this: subscribe to every AI product you can find.
I would not recommend that. A pile of subscriptions can easily make a solo business worse. Every new surface creates another place where work can disappear, another place carrying stale context, another set of files and another bill. If every provider becomes its own isolated version of the company, more AI can create more fragmentation instead of more leverage.
The reason Eterna can use this many systems comfortably is increasingly because each one has a bounded role and the company does not live inside any one of them.
If I were starting again, I would still begin with one strong daily-driver AI and the ordinary systems the business actually requires. I would add another provider when repeated use showed that it genuinely handled a class of work better. I would keep accepted files and important state outside the conversations and use ordinary software whenever a task became exact enough that repeated reasoning was unnecessary.
Most importantly, I would pay attention to what I was learning repeatedly. If I solve the same process five times, perhaps it should become a workflow. If I explain the same standard repeatedly, perhaps it should become durable guidance. If a model keeps performing the same mechanical transformation, perhaps that transformation belongs in code. If one reviewer catches the same failure repeatedly, perhaps that failure needs a test.
That is how more tools can eventually create less work.
The leverage is becoming cumulative
This is the part I find hardest to compare with the way I worked before Eterna. The biggest benefit is no longer simply that an AI can help me do something faster today. It is that today's work can reduce how much work the next version requires.
A provider helps solve a hard problem. Part of the solution becomes deterministic software. A process becomes a playbook. The next provider can retrieve the playbook. A failure becomes an evaluation. A useful pattern becomes part of EternaAI's training. The Workspace makes the resulting capability available again without requiring me to remember exactly which chat originally figured it out.
That changes the economics over time. The subscription may cost roughly the same next month, but the system using the subscription can be better.
I think that is the real opportunity for a solo founder now. It is not simply that AI lets one person type faster or produce more content. One person can assemble an unusually broad set of specialised capabilities, connect them to ordinary software, preserve what works and gradually turn repeated intelligence into operating infrastructure.
I still find new tools useful, and I still get excited when a provider releases something genuinely better. But I am becoming much less interested in finding the one system that does everything. The stack I already have is getting more useful every time Eterna learns how to use it better.
Every time I look at what is now running through that stack and then look at the monthly software bill, that is still the part I have trouble getting used to.
Top comments (0)