Are we seriously normalising the absurdity of shipping personal data, company documents, and sensitive content to remote AI servers — even when we don’t need to — for processing we could have handled locally all along?
I’ve been reading several interesting pieces on Local AI lately, and they make painfully clear the sheer scale of the colossal blunder we’re committing by not adopting a ’local first’ policy for every AI application we build.
The problem: cloud everywhere, common sense nowhere
Today we bolt OpenAI or Anthropic APIs onto everything, often for tasks as mundane as:
- Summarising notes
- Classifying documents
- Organising emails
- Transforming text in general
But to perform these relatively simple operations, we send data off our own devices. And in doing so, we create:
- Pointless cloud dependencies
- Higher costs and latency (data has to travel from us to the cloud and back)
- New failure modes (what happens when the network drops or GPT doesn’t respond?)
- Privacy and compliance headaches
- A brittle operational stack
Have we completely lost the plot?
You already have the compute power — did you know that?
Modern smartphones and laptops already pack enormous onboard AI capability: dedicated GPUs and often enough RAM to run near-frontier models (properly quantised). All that powerful hardware sitting largely idle.
If the vast majority of AI features don’t require the superintelligence of a frontier model and simply need to transform local data reliably, why aren’t we reaching for a good SLM (small language model)?
This approach makes far more sense, because SLMs:
- Are faster
- Cost nothing beyond the energy to run them
- Work offline
- Are more predictable
- Offer stronger privacy guarantees
The story that should have told us everything
Did you know there’s someone who took a 10-hour flight with an SLM-enabled Mac and spent the journey vibe-coding a billing analysis tool — entirely in the air?
How good is that? And how much more freedom and privacy does that give you?
(Read it here: https://lnkd.in/dZXj9fQf)
That’s the path. That’s the kind of freedom we should be chasing.
Two completely different roads
Over the next few years, I think we’ll see two very distinct directions emerge:
- Cloud AI for complex reasoning and global knowledge
- Local AI for productivity, personal automation, and data transformation
And the second category will probably grow far faster than anyone expects — and faster than the first — because, just as it was forty years ago with the shift from mainframes to personal computing, a tool’s full flexibility only reveals itself when you’re free to use it on your own terms: without the constraints of forced sharing, and without sending data across the wire when there’s no good reason to.