We’re closer than you might think to a world where a personalized AI assistant lives in your home. At TechCrunch Disrupt 2026, Andrew Feldman, CEO and co-founder of Cerebras Systems, sketched out a practical vision: compact, appliance-like AI agents that run primarily on local devices but lean on the cloud when needed. For Feldman, the pieces are already falling into place — consumer-facing agents from major players and startups show people want assistive AI, and the underlying hardware is catching up to make it feasible for those agents to live in our homes.
Why home AI feels imminent
Feldman pointed to recent consumer agent efforts — such as Meta’s Muse and other commercial products — and to well-funded startups that are demonstrating broad interest and momentum. Those examples suggest users will accept agentic software that proactively helps with everyday tasks. Because chipmakers, software platforms and cloud services are rapidly improving, the concept of an “AI appliance” is no longer science fiction: it’s an engineering and business problem that’s now solvable.
How personal AI assistants might operate
Rather than a single monolithic system, Feldman envisions a hybrid architecture: devices that do “some work on device and some work going back to the cloud.” In practice this means:
- Latency-sensitive or private tasks handled locally (on an in-home appliance or endpoint).
- Heavy model inference or aggregated updates performed in the cloud.
- Seamless switching so the assistant feels instantaneous for the user while retaining the benefits of large-scale models.
This split is the same idea behind modern smartphones — local processing for routine interactions, cloud resources for complex computation — applied to always-on, personalized agents.
Trust, security and the new perimeter
A core part of Feldman’s prediction is that these assistants will “get to know you very well” and operate within a defined security perimeter. That raises immediate questions about safety and trust: if an assistant has access to payment methods and can make purchases, how is that access controlled? How will identity, authentication, and consent be managed at the device level? Feldman’s framing implies that designers will need to build robust, user-centric security models so that convenience does not come at the expense of control.
Real use today: agents as productivity multipliers
Feldman isn’t speaking purely as a theorist. He described using “a collection of agents” to automate practical tasks: sweeping his calendar, preparing context for meetings, and surfacing information about people he’ll speak with. Those day-to-day productivity gains hint at the low-hanging fruit for consumer adoption: savings in time, better organization, and context-aware assistance that augments what humans already do.
Why infrastructure builders are excited
As the CEO of a chipmaker, Feldman naturally appreciates the hardware opportunity. But he also made clear that infrastructure builders don’t always need to invent the consumer-facing application — they can focus on platforms and tools that enable others to create the services people will use. If personal AI at home becomes mainstream, demand for specialized chips, efficient local inference, and secure device fleets will surge, creating a rich ecosystem for startups and incumbents alike.
Potential societal trade-offs and the road ahead
Widespread personal AI appliances bring benefits — convenience, personalization, and new capabilities — but they also force trade-offs. Privacy expectations will evolve when assistants store and process sensitive data locally or sync it to the cloud. Regulatory scrutiny and consumer education will matter. Designers will need transparent defaults, clear consent flows, and fail-safes for financial or critical actions.
What to watch next
If Feldman’s timeline is correct, the next 12–36 months should reveal:
- More consumer-grade agent launches from big tech and startups.
- Hardware advances making local inference cheaper and faster.
- Early standards or best practices for device-level security and payment authorization.
- An uptick in third-party services building on local-first agent platforms.
The idea of a household AI assistant shifts from novelty to inevitability as technical and market forces align. Feldman’s perspective — grounded in both product use and infrastructure strategy — suggests that a future where your home has an intelligent, helpful companion isn’t just possible; it’s becoming practical.
Licensed under CC BY-ND 4.0 International
