Introducing model profiles

Not every task needs the same model. Sorting a morning of support tickets should be quick. Reading a forty-page spec and comparing it with last month's version needs depth. Until now, every workflow in Runlance used the same setting, which meant teams paid for depth they did not need or waited longer than necessary for simple results.

Model profiles change that. Released in beta with version 1.4.0, they let you choose how much effort an agent spends on each workflow, with relative speed and cost always visible.

Three profiles, one setting

Each workflow now has a default profile. You pick it when you create the workflow and can change it at any time. There are three options:

• Quick: the fastest and lightest option. Good for triage, short summaries and tasks where the sources are small and well structured.

• Balanced: the new default for most workflows. Good for weekly briefings, sprint summaries and account briefs.

• Deep: the most thorough option. Good for long documents, comparisons across many sources and drafts where missing a detail is costly.

Next to each profile, Runlance shows relative speed and cost on a simple scale, so the trade-off is clear before you save. Profiles are named by what they do, not by the model behind them, so your workflows stay the same when we update the underlying models.

How to choose a profile

A simple rule works for most teams: start with balanced, then adjust based on the review history. If reviewers approve almost every run without changes and the task is simple, try quick. If reviewers keep asking for missed details, try deep.

A few questions help when you are unsure:

1. How long are the sources the agent reads?

2. How costly is a missed detail in the draft?

3. How quickly does the result need to be ready?

4. How often does the workflow run?

A workflow that runs every hour on short tickets is a good fit for quick. A monthly comparison of contract notes is a good fit for deep. Most weekly briefings sit comfortably in balanced.

Choose the profile for the task, not for the team. One team can run quick and deep workflows side by side.

Profiles and review

Profiles do not change how review works. Every run still pauses for a named reviewer, every answer still cites its sources and every run still appears in the activity log with the profile it used. That makes it easy to compare. Replay a run with a different profile and look at the two drafts next to each other before you change the default.

We recommend changing one thing at a time. If you switch the profile and add a new source in the same week, you will not know which change improved or hurt the draft.

Profiles also do not change permissions. A deep profile reads more carefully, not more widely. It still sees only the sources and fields you allowed for that workflow, and every source it used is listed in the activity log. If a draft needs information from a new place, add that source on purpose, rather than hoping a stronger profile will find it.

What we are still learning

Model profiles are in beta. During early testing, we saw teams move simple triage workflows to quick and keep their weekly briefings on balanced, which matched what we expected. We also saw a few teams choose deep for everything, which made runs slower without improving drafts that were already approved as they were. We will keep refining the relative speed and cost indicators as more workflows use them.

If a profile does not fit your task, tell us in the feedback panel on any run. Notes from reviewers are the most useful input we get.

Getting started

Open any workflow, go to its settings and choose a profile. Existing workflows start on balanced, so nothing changes until you decide it should. New workflows ask for a profile when you create them. A good first step is to review your three busiest workflows and ask whether each one really needs the depth it has today.