Product
AI Email Summaries Are Not Enough
Compression is useful. It is not the same thing as judgment.
AI made summarization cheap very quickly.
A long thread can become five bullets. A morning inbox can become a paragraph. A week of messages can become a neat brief before you finish coffee.
That is genuinely useful.
It still leaves the harder problem untouched.
The wrong thread can be summarized perfectly
A summary answers: what happened here?
Attention requires another decision first: which “here” deserves me at all?
The most consequential conversation may not be the longest, newest, or most active. It may be the pricing question you already opened. The quiet introduction that suddenly became relevant. The thread where someone promised to return with a decision and never did.
A better summary of the wrong threads is still the wrong allocation of attention.
Summaries are retrospective
Most summaries begin with a set of messages and compress them.
A radar begins with the opposite problem. It has too many possible messages and must decide which few are worth surfacing.
That requires state, history, timing, and some model of consequence. It also requires the ability to return nothing when nothing deserves interruption.
Why this matters for agents
As models get better at generating text, generation becomes less differentiating.
The scarce capability moves upstream: selecting the right context, recognizing the right moment, and deciding whether an action is warranted.
This is the direction Sidekick Log is moving toward. Drafting is useful. Summaries are useful. But the more fundamental product question is whether the system can make a better decision about what deserves attention before any text gets generated.
More intelligence does not automatically create better judgment.