Proactive systems: attention, timing, and feedback

I once thought memory was the most important capability of a personal system intended to last for years.

If it remembered my goals, notes, preferences, and past decisions, it would become more useful over time. We let information enter a knowledge base, allowed conversations to leave durable traces, and gave later answers access to more context.

All of that mattered. It also exposed a more immediate problem: if I still had to ask first every time, the system remained a question-answering tool with a better memory.

A true personal system should not always wait to be asked.

Memory does not create initiative

A system can preserve many facts about a person without knowing when any of them should be used.

The notes may contain a long-running interest. The task system may contain a clear goal. The outside world keeps producing new changes. Unless the user sends a request, these sources remain silent beside one another. Memory makes the past retrievable. It does not automatically make the present meaningful.

Initiative requires another capability: detecting that a change relates to a current goal and deciding whether it deserves the user’s attention.

This is not merely retrieval. Retrieval asks where the information relevant to a given question is. A proactive system must first decide whether a question has appeared at all.

A schedule is not proactive intelligence

The easiest proactive system to build is a cron job.

Send a report at a fixed time every day. Check a status at regular intervals. Remind the user when a task reaches its deadline. The product acts without a prompt for the first time, which makes scheduling easy to confuse with proactive intelligence.

The arrival of a scheduled time does not mean there is something worth saying.

A fixed report may contain no new information for several days. A reminder may continue after the user has already acted. The system technically completes its task while repeatedly taxing attention with the same content.

We encountered excessive reminders, duplicate messages about the same event, and updates sent mostly to prove that the system was still working. That changed how I measured initiative. The number of messages a system produces is not evidence that it is proactive in a useful way.

The most important capability of a proactive system may be knowing when to remain silent.

Proactivity needs an attention contract

Once a system can interrupt a person, it gains a permission more sensitive than generating content: it can allocate attention.

That permission needs an explicit contract. Every signal should pass through several judgments:

• Relevance: does it relate to a current goal or responsibility?
• Novelty: has anything actually changed from what the system already knows?
• Actionability: can seeing it now lead to a different choice?
• Urgency: will the value disappear if it is seen later?
• Confidence: is the evidence strong enough to justify an interruption?

Not all information should become a notification. Low-value material can be stored. Relevant but non-urgent changes can enter a digest. Only changes with both timing and decision value should interrupt the user immediately.

A personal system therefore needs at least three distinct actions: remember, summarize, and interrupt. Implementing all three as instant messages is the simplest and noisiest design.

Context is not the complete personal archive

To improve proactive judgment, a system naturally wants more information: notes, calendars, tasks, long-term goals, and whatever the user has recently been following.

Access to more material does not mean every judgment should receive all of it. A person’s history is vast. Their present situation is always local.

Useful context answers a narrower set of questions: what state is the user in now, what remains unresolved, which constraints still apply, and which decision would this new signal change?

This is why a knowledge base cannot remain only a growing folder. Notes need roles. Some are stable facts, some are hypotheses under examination, some have expired, and some belong only to one temporary action. If the system cannot distinguish them, remembering more makes proactive judgment increasingly vulnerable to stale information.

Personalization is not knowing everything about the user. It is using the right small part at the right time.

A proactive system must observe outcomes

The work does not end when the system sends a notification.

Whether the user reads it, acts on it, and eventually benefits from it determines whether the interruption had value. If those results never return, the system can only continue sending according to its original rules. It cannot learn whether its judgment was correct.

A real loop looks like this:

1. Internal or external state changes.
2. The system evaluates whether the change affects a current goal.
3. It chooses to store, summarize, ask, or interrupt.
4. It observes the user’s action and the later outcome.
5. It adjusts the next trigger and the way it communicates.

The final step turns proactive automation into a learning system. It does not only execute predefined rules. It gradually learns what matters to this particular person.

Personality is not judgment

Personal systems easily acquire a distinctive voice, recurring forms of address, and language that feels familiar with the user. Personality creates continuity, but it can also hide the fact that the system is not making better decisions.

A warmer reminder can still be useless. A confident report may simply present uncertain information with greater conviction.

Long-term trust comes from respecting attention, admitting uncertainty, and changing behavior after mistakes. Personality can support the relationship. It cannot replace evidence, boundaries, or feedback.

A personal OS is a decision system

I no longer think of a personal OS as a place that contains notes, tasks, and conversations. Storage is only its most basic capability.

The system should reduce the distance between signal, judgment, action, and outcome. It should know what deserves to be remembered, what should be summarized later, what must appear now, and when it should say nothing.

A passive tool waits for input and then provides an answer.

A proactive system keeps observing while treating attention with restraint.

Its value is not how much it tells you every day. It is that when something important happens, you do not have to know in advance what question to ask.