Two AI models went dark worldwide this summer. Nobody got a warning.
A list of models is a dropdown. The layer underneath is the product: it routes the task to the model that fits and keeps your context when one provider drops.
That layer is what we built.
Two AI models went dark worldwide this summer. Nobody got a warning.
A list of models is a dropdown. The layer underneath is the product: it routes the task to the model that fits and keeps your context when one provider drops.
That layer is what we built.
@ashwingop "Rent the intelligence, own the context" is the architectural bet. We built Swa around exactly this: model-neutral context layer, platform-neutral surface (Slack, Teams, WhatsApp, SMS, web). Intelligence rotates. Operating memory stays yours.
@karpathy@salomon_diei "Everyone is a manager" is the right framing. Genuine q though: when a marketing team asks Claude Tag for an AI-generated image, what happens? Single-model setups seem to need a workaround every time a request falls outside Claude's native skills.
97% of AI breaches happen where there are zero access controls.
Your engineers are already routing around your infrastructure to use the tools that work.
That is not recklessness. That is a design failure.
Full story tomorrow. Thread below.
@BenPTweet This is exactly why we built Swa.
When the approved platform fails engineers, they find another way. The breach follows.
Zero data retention by default. Your models. Your rules. → https://t.co/uh47qv5TvK
@BenPTweet This is why we built Swa. Enterprise AI inside Slack and Teams , no new tab, no workaround, zero data retention. The governed path has to be the easy path.
Machine speed. Human response time.
That gap isn't theoretical anymore.
Anthropic's Mythos found a 27-year-old Linux kernel bug that automated scanners missed millions of times.
It found it in minutes. Autonomously.
Thread on what this actually means for enterprise leaders
#AIGovernance #EnterpriseAI