Familiar work, answered. New work, briefed.
MS Duplex answers what it has learned in milliseconds, briefs your AI for what is new, and learns from every confirmed result.
One request, every way it can go.
As the Agent Operating System your agents run on, MS Duplex takes part in every request: before it, at each model and tool call, and after the result.
Step 1 of 7: Before day one
Answered in milliseconds.
Familiar work skips the model call, so it comes back faster and costs less.
Step 1 of 3: Time to decide
A decision MS Duplex has learned is answered in ≈97 ms, where a model call takes ≈2,231 ms.*
What counts as confirmed.
Each work area sets its own standard, and MS Duplex learns only from results that meet it.
Step 1 of 4: A passed test
When a result meets the test your team chose, it is confirmed, and MS Duplex learns from it.
What it learns stays yours to correct.
What MS Duplex learns from, what it leaves out, and how you change it.
Step 1 of 4: Learns from
MS Duplex learns from pretraining material, confirmed corrections, its decisions and their outcomes.
Confidence and autonomy.
MS Duplex knows when it knows, and hands over the rest.
Step 1 of 5: Known work
A repeat quote is one MS Duplex has learned, so it answers at once, with no model call.
