Blog / WEEKLY #005
#005 Holiday catch-up: the personal agent race heats up; will games, software and hardware all go open source?
Flash off-the-shelf firmware, rework the interface into Chinese, bolt on a voice output path, and a small desktop dev board goes from toy to an assistant terminal that gets used every day; as for taking over the whole house, that is still just a vision.
Recording by itself produces no value; review does. When watch, training, and health data are automatically compiled into a weekly report, continuous logging and scheduled review finally connect into a closed loop.
Phone performance is already more than enough, yet an Agent can barely access anything on it: from permissions and sandboxes to app isolation, the mobile bottleneck lies beyond raw compute.
As the cost of generating content approaches zero, what becomes scarce is no longer output but a reason for people to keep coming back. Works, a consistent set of preferences, and feedback from real users are where trust comes from.
A Bluetooth remote costing a few dozen yuan plus a small voice program written at home made for two straight days of fully voice-driven work: how a physical button changes what it feels like to give an Agent its orders.
Switching to open-source or self-trained models to cut costs only works once the business is stable and quality can be verified independently. This article discusses the conditions under which that decision should be made.
Moving plugins from one agent to another, format conversion is no longer the hard part; the real threshold is that each platform exposes different capabilities, and upstream keeps changing fast.
As models start acting on intent rather than literal instructions, the center of gravity in collaboration shifts from prescribing every step to stating goals and acceptance criteria. Which constraints still need to be written down explicitly is worth discussing separately.
The rendered layer and the source are two separate states, and that is the root of a whole class of stubborn problems in WYSIWYG editors. One migration from Milkdown to CodeMirror 6 shows where the openings for modifying an existing system lie, and who owns the responsibility for acceptance.
A game script and worldbuilding document that sat on a hard drive for a year became a playable experience prototype over this holiday. Which stretch of the road from idea to playable got compressed, and which judgments remain in human hands.
How open a piece of software is to agents depends on whether it exposes capability interfaces, whether results can be inspected, and whether the host can carry the execution. From 3D modeling to editing tools, this line of judgment is taking shape.
Projects, caches, system images, and raw video fill the drive together. The question is not only how much capacity to buy, but which data is worth keeping, on which tier, and whether it can be found again when needed.
Looking things up, placing orders, paying, making promises to others — delegating to an agent involves different tiers of authority. The difference between an assistant and a digital alter ego hides in every single "should I ask first?"
Whichever service holds your memory is the one you have to ask for it back when you switch tools. A file system you own lets personal context move between multiple agents — but sync, backup, and multimodal material are still unfinished roads.
The ceiling of what a life assistant can do often is not the model — it is where the information it can see lives: the same school notice is readily at hand in email, but hard to reach inside a group chat.
Personal agents have lowered the bar to "being able to chat", but what decides retention is not the chatting — it is the first useful task: scenario entry points, an execution process you can see, and a conversation the user never has to manage.
Resale discounts on model APIs keep thinning; profit has to be earned by turning tokens into usable services. This article maps the cost structure of AI applications and the opportunity in "experience arbitrage."
Whether logins, software, and files survive after a task ends is splitting agent execution environments into two generations. Persistent environments make scheduled tasks and long-running collaboration possible — and push cost and multi-machine management onto the user.
When content serves both human readers and agents, the open web and a subscribable entry point take on new value. A gap discovered while assembling AI sources shows this still has to be done deliberately on the Chinese internet.
