26 July '26 | 
M.J. Schermer
Should you start a new project from scratch, with Python? Or even with a framework?
Well, you could, but most Python frameworks make workflows harder to change later.
Should you use an agentic harness?
Well you could, but agentic harnesses make workflows harder to control + more expensive to run.
The overall result is the same for both paths: Loss of control over your AI usage.
So what if you kept your AI workflows simple, for example using a special yet simple language?
What if simplicity indeed is truly "the ultimate sophistication"?
What if simplicity indeed leads you to waste less tokens, time and money?
Contrary to popular belief, prompt engineering is not where most of your token waste is in the long-term.
If you're looking for a short-term solution, you can use prompts like:
"Keep the conversation natural and short, no more than one paragraph, no more than 30 words:"
Sure you can save some tokens here.
However, sooner or later you will need more data and more workflow steps.
I started using Nyno, because I already tried to do custom integrations in multiple programming languages. Spoiler alert: That does not scale well.
I spend weeks connecting OpenAI to a CMS project, only to realize that for European companies Mistral AI might be the better choice.
The overall problem is that you keep rewriting and rewiring your AI capabilities in many different places and languages.
As a result, you lose a lot of time, and if you are not losing time you are losing control or flexibility.
Instead now, with Nyno, you can build something that works for any project, in less than 10 minutes, because it's not deeply dependent on each of your projects.
You can ONE-SHOT AI with:
Nyno
workflow:
- step: ai-mistral-text
args: ['your prompt']Or use past history as well:
context:
MISTRAL_MESSAGES: [..] # past messages
workflow:
- step: ai-mistral-text
args: ['your prompt']Or even past history with more steps:
context:
MISTRAL_MESSAGES: [..] # past messages
workflow:
- step: ai-mistral-text
args: ['your prompt']
workflow:
- step: ..Notice how short these workflows are?
We even have a visual YAML workflow builder and you can directly use the YAML files as password protected commercial APIs (without needing a license like with n8n!):
GitHub Link: https://github.com/flowagi-eu/nyno
Even if you are not a Nyno fan (yet), or want to build your own (compatible) YAML solution, the overall guiding success principle for wasting less AI tokens in the long-term is the same:
Avoid making programming languages your AI Core language.
Simplicity is the ultimate sophistication.