AI planning

AI that estimates, infers, and asks before acting.

Two AI features pull their weight in Narcove. Soundings gives you honest time guesses — low, likely, high — based on your past work. Plan-with-AI reads a brief and proposes a schedule with dependencies. Both show you a diff before anything changes. You stay in control.

Try AI planning for freeOther features

What's hard right now

Two parts of planning that nobody likes

Time guesses are theatre. Someone asks how long a card will take. You guess. Maybe two days. Maybe three. You don't know. You write down a number to be helpful. You and the team quietly agree it's a guess. You ship late, but everyone's sort of OK with it because nobody believed the number anyway.

Building a schedule from a brief takes hours. A new project starts. You read the brief. You write cards. You add dates. You add dependencies. You move cards. You add more dates. By the time the schedule looks right, half a day is gone, and the brief has changed.

Soundings

Time guesses as ranges, not single numbers

Every card gets a low / likely / high guess based on cards your team has already shipped. The likely number is what most similar work took. The range is where 80% of your past work fell. Schedules built on ranges flex when reality does.

  • Why-line every time. One sentence on which past cards anchored the guess. You see the working.
  • Bulk estimate. Estimate every unestimated card with one click. Useful at the start of a project.
  • Override anything. Soundings is a starting point. Type your own number and the AI guess becomes a sanity check.
  • Tightens with use. First 30 cards are wide. After 100 they're close to your team's real speed.
A card receives a guess — low 3 days, likely 4, high 5 — with a one-line reason: based on three past brand projects.

How Soundings work — five steps

  1. 1

    Open a card

    Click any unestimated card. Hit the Sparkles icon next to the duration field.

  2. 2

    We read your past work

    We pull cards with similar titles, tags, or descriptions from your project. We weight by how recent they were and how long they actually took.

  3. 3

    We propose a range

    You get a low / likely / high guess in minutes, hours, or days. The likely figure is what most similar work took. The range is where 80% of your past work fell.

  4. 4

    You see the why

    A one-line note shows which past cards anchored the guess. "Based on three logo redesigns and one icon set." You see the working, not just the answer.

  5. 5

    Accept, edit, or override

    One click accepts the range. You can edit any of the three numbers. You can also override the whole thing with your own value — the AI guess stays visible as a sanity check.

Plan with AI

From a backlog to a working schedule

Drop in a list of cards. Claude reads them, infers which one needs which, checks the graph for cycles, and lays out a Chart that respects working hours and bank holidays. You see the diff before anything changes.

  • Inferred lines. Dependencies suggested from card content, not an empty graph.
  • Cycle-safe. Database-level check refuses loops. You can't accidentally apply a broken plan.
  • Resource-levelled. Two cards on the same person never overlap.
  • Diff before apply. You see the proposed Chart next to the current one. Accept all or pick what to apply.
A small dependency graph: research feeds wireframes and copy, which feed visual design, build, and launch.

How Plan-with-AI works — six steps

  1. 1

    Drop in a brief or backlog

    Paste a brief, a feature spec, or a list of cards. Or import a CSV from any tool.

  2. 2

    Claude reads each card

    Title, description, any tags or labels you've added. We pass the lot to Claude with structured tool calls.

  3. 3

    Dependencies get inferred

    Cards that reference each other get a suggested dependency line. "Wireframes depends on research." You see the graph as a Gantt.

  4. 4

    Cycles get refused

    If the AI suggests a graph with a loop, the database refuses it at the trigger level. You see a clean graph or no graph — never a broken one.

  5. 5

    Schedule respects working time

    Each card gets a start and finish based on dependencies, your working hours, your team's hours, and bank holidays. Two cards on the same person never overlap.

  6. 6

    Review the diff

    You see the proposed Chart side-by-side with the current one. Accept the whole thing or pick the cards you want to apply.

Use cases

Four times AI planning saves an afternoon

Why this matters

AI that helps without taking over

Most AI in PM tools either does too little or too much. The chatbots that draft emails are fine but don't change how you plan. The autopilots that move cards on their own feel like they're going to break something — and sometimes do. Both miss the point. The hard part of planning is the guesswork, not the typing.

Soundings and Plan-with-AI tackle the guesswork. They take the parts of planning that everybody hates — making up time numbers, building a graph from a brief — and propose answers grounded in your team's actual data. You still review. You still decide. But you stop starting from a blank page and you stop pretending you know how long things take.

The guardrails matter. We never auto-apply a plan. We always show a diff. We always cite past cards in the why-line. We stay opinionated about what AI is good at — pattern matching across a lot of past work — and what it isn't — making business decisions for you.

FAQ

AI questions

Which AI model do you use?

Anthropic's Claude. We chose it because it's good at structured tool calls — that means the AI fills in fields directly instead of writing JSON we have to parse. The result is far less likely to hallucinate or output something we can't use.

Can the AI move my cards without asking?

Only the Tide auto-applies safe changes — pull-forwards and small non-critical push-backs. Soundings and Plan-with-AI always show you a diff first. You see what would change. Nothing actually changes until you click apply.

Where do the time guesses come from?

Soundings reads your project's past work. Card titles, descriptions, tags, durations, and how long similar cards took to finish. The first guesses are wide because we don't know your team yet. After 30 cards the ranges tighten. After 100 they're close to your real speed.

Do you train on my data?

No. Your cards stay in your workspace. We send the relevant card data to Claude for one inference at a time, then drop it. We don't train, fine-tune, or retain anything beyond what you can see in your own logs.

Is there a usage cap?

Yes. Pro includes 200 estimations and 50 plan runs per seat per month. Team includes 600 and 200. Past the cap you get read-only suggestions until reset. Most teams never hit it.

What if my work is brand new and has no past data?

We start with a wide range based on similar cards in our anonymous baseline. Your first 30 cards tighten it to your team's speed. The early ranges are wider on purpose — they're honest about what we don't yet know.

Can I override an AI guess?

Always. Soundings is a starting point, not a verdict. Type your own number and the AI guess becomes a sanity check next to it. Override often enough and the AI learns from your overrides on the next round.

How does Plan-with-AI handle dependencies?

It reads card titles and descriptions, looks for words like "requires", "after", "based on", and infers a dependency graph. The graph is checked against the database — cycles are refused at the trigger level — so you never get a plan that loops. You see the suggested graph and can edit it before applying.

Related

How AI planning fits with the rest of Narcove

Try AI planning for free

Pro and Team include AI features. Free tier has the kanban and portal without AI.