Blog / Engineering
How we think about AI in your workflow
Suggestions, not substitutions—what we optimize for when models meet real meeting notes.
By Glen Conway, Founder · · 2 min read
The job is not “generate text”
Meeting transcripts are noisy. People interrupt, revisit topics, and sometimes disagree without resolving. A useful AI workflow does not pretend the transcript is a spec. It proposes structure you can correct.
That is why we bias toward:
- Extracting candidate actions with owners and dates when the language supports it
- Leaving ambiguity visible instead of inventing precision
- Letting you accept, edit, or discard before anything becomes team-visible
Code you can trust: keep humans in the merge path
Internally, we treat model output like a diff. The same instinct applies in product: nothing should silently become “official” without a human click.
// Illustrative: proposals are staged until explicitly applied
type ActionProposal = {
summary: string
assignees: string[]
targetDate?: string
confidence: 'high' | 'medium' | 'low'
}
function applyProposal(p: ActionProposal, approved: boolean) {
if (!approved) return
// persist as a real action…
}
Privacy and proportionality
We design features so that the sensitive payload—your transcript—is handled with clear boundaries, and so that you can work without oversharing. If you are evaluating any AI tool, ask two questions:
- What is retained, and for how long?
- Can you export or delete your workspace data on demand?
When AI helps most
The wins we see consistently:
- Long calls with many speakers and overlapping threads
- Recurring meetings where similar action types appear every week
- Handoffs where names and dates are spoken but not typed into a tracker yet
Pull quote
If the AI saves ten minutes but creates ten new “is that actually what we agreed?” messages, it failed.
Related reading on this blog
- Closing the loop on meeting action items — why follow-up drifts between calls.
- A practical checklist for status calls — a lightweight ritual you can reuse weekly.
- New to Action Log? Start from the home page for a product overview.
Closing thought
AI in Action Log is a drafting partner, not an author. The goal is fewer hours spent re-typing commitments—and more time making sure the right commitments ship.