Skip to content
Advancing Bits to Master Atoms

Long-horizon planning.
Weeks-long campaigns stay one job, with plan and progress.

MS Duplex keeps a long campaign as one job, with its goal, its route and the evidence gathered at every checkpoint. Give it the goal rather than a script, and it plans again when a result changes the picture.

Plan and decide · one of the fourteen capabilities of MS Duplex

A winding track of graphite tiles climbing a terraced graphite form, with slim silver flags at intervals and a cobalt-blue flag at the summit.A winding track of graphite tiles climbing a terraced graphite form, with slim silver flags at intervals and a cobalt-blue flag at the summit.

Your team gets

  • Multi-step research that reaches its goal
  • Checkpoints your experts can review at any time
  • Work that survives interruptions and hand-offs

What long-horizon planning means for your team.

Long-horizon planning means your AI can carry a goal that takes many steps to reach. MS Duplex lays out the route from where the work stands to where your team wants it to be, keeps a checkpoint at each stage and records the evidence gathered there.

An agent on its own works one session at a time. When the session ends, the plan, the progress and the reasons behind each step end with it, and the next session starts by rebuilding them. MS Duplex keeps the campaign itself, so a pause, a hand-off or a new session picks up from the last checkpoint.

A plan is only useful while it still fits the facts. When a result comes back that changes the picture, MS Duplex plans again from that point, keeps what still holds and records why the route changed.

Why research needs plans that last.

Real research is rarely one step. The goal is reached through a series of runs, analyses and reviews, and each one shapes the next.

A campaign to reach a target, compare a set of designs or close an open question can involve dozens of steps, several people and more than one tool. The value lies in the sequence: what was tried, what came back and why the next step was chosen.

When that sequence lives only in a chat window or in someone’s head, every interruption costs the work. A weekend, a colleague’s leave, a model change or a closed session means the plan is rebuilt from memory, steps are repeated and the reasons behind earlier choices are lost.

  • Goals take many steps

    The questions that matter most to a research team are answered through a campaign of runs, rarely by a single one.

  • Work is interrupted

    Sessions end, people hand over and priorities shift, and the campaign has to carry on through all of it.

  • Results change the route

    An unexpected result should change the plan from that point on, while everything that still holds is kept.

  • Leaders need to see progress

    Anyone responsible for a campaign should be able to see where it stands and what each step rests on.

A satin-silver plate on a graphite base, a few silver pins standing on it and one thin cobalt-blue line routed from pin to pin.A satin-silver plate on a graphite base, a few silver pins standing on it and one thin cobalt-blue line routed from pin to pin.

How long-horizon planning works in your team’s day.

Your team sets the goal and reviews the checkpoints. MS Duplex keeps the plan, the progress and the evidence between them.

  1. 1

    The goal is set

    A researcher gives the agent the goal and the constraints that matter, such as the budget and the steps that need a person’s approval.

  2. 2

    The route is laid out

    MS Duplex lays out the steps from where the work stands to the goal, with a checkpoint at each stage your experts can review.

  3. 3

    Each checkpoint is kept

    As the agent works, the result and the evidence at each checkpoint are recorded, so the progress is on record at every stage.

  4. 4

    The plan adapts

    When a result changes the picture, MS Duplex plans again from that point and records why. A new session, agent or person picks up from the last checkpoint.

An example, in any field where an agent runs a campaign.

A team’s agent is asked to find the best of a large set of candidate designs within a fixed budget of runs. The campaign spans several sessions and three people.

  1. 1

    The plan

    The agent receives the goal and the budget. MS Duplex lays out a route: screen the full set, narrow it to a shortlist, then run the detailed evaluation on the shortlist, with a checkpoint after each stage.

  2. 2

    A result changes the picture

    The screening shows that one group of candidates fails a requirement the team cares about. MS Duplex plans again from that checkpoint, drops that group and moves the budget it would have used to the shortlist, recording why.

  3. 3

    A hand-off

    The researcher who started the campaign moves to other work. A colleague opens a new session on a different model and starts from the last checkpoint, with the goal, the decisions and the evidence in view.

  4. 4

    The goal is reached

    The detailed evaluation confirms a leading design. The team lead reviews the checkpoints and sees every step that led there and what each one rested on.

Long-horizon planning works with the other capabilities.

A long campaign draws on most of the layer, and every step it takes feeds the rest.

  • Simulation before action

    Before each costly step in the plan, the options are played forward, so the budget goes to the runs most likely to move the campaign on.

  • Research memory across sessions

    Every new session in the campaign starts from its current state: the goal, the decisions, the approved methods and the progress.

  • Rules and approvals

    In governed environments, the steps you select, such as an expensive run, wait at their checkpoint for a person’s approval.

  • Continual learning

    What each checkpoint confirms is learned at once, so the rest of the campaign, and the next one, starts from it.

What your team gets.

Long research that reaches its goal, with its plan and progress in view at every stage.

  • Multi-step research that reaches its goal

    The agent carries a goal across many steps and plans again when the facts change, until the work is done.

  • Checkpoints to review

    Your experts can review the plan, the progress and the evidence at any checkpoint, at any time.

  • Work that survives interruptions

    A pause, a weekend or a new session picks up from the last checkpoint, with the work intact.

  • Clean hand-offs

    A colleague, or a colleague’s agent, takes over a campaign from where it stands, with every decision and its reason in view.

  • Plans that follow the evidence

    When a result changes the picture, the route changes from that point and the reason is recorded.

  • One record per campaign

    Work that spans people, tools and sessions stays one campaign with one plan and one record.

Common questions.

What is the difference between a plan and a script?

A script fixes every step in advance. A plan in MS Duplex starts from the goal, lays out a route and changes from the point where a result changes the picture, keeping everything that still holds.

What happens when a session ends in the middle of a campaign?

The campaign carries on from its last checkpoint. The next session, on the same agent or another one, starts with the goal, the decisions, the evidence and the progress so far.

Can our experts step in during a campaign?

Yes. Every checkpoint can be reviewed at any time, and an expert can correct a step or change the goal. In governed environments, the steps you select wait for approval before they run.

Does long-horizon planning work in our field?

It works in any field where a team reaches a goal through a series of runs, analyses or reviews with AI. The plan is built around your goal, your tools and the results your team confirms.

Make the next run benefit from the last.

Bring us the research your AI should get better at. We will plan a pilot on your own workflows, with your experts as the judges.