All topics

Planning sessions

Pressure-test a change plan before OrgBolt builds it. Type /planning and the AI asks one grounded question at a time — about the automation, access, and data your change actually touches — then writes up the decisions and risks so the build follows them.

What a planning session is

Most Salesforce changes fail on the things nobody thought to ask: the flow that already writes the same field, the profile that can't see it, the historic records a new rule would block. A planning session front-loads those questions. Before anything is built, the AI reads your org's real metadata and walks through what your plan touches, one question per turn.

Questions come with quick-answer buttons when the answer is a genuine either-or (press 1–9 to pick), and you can always type a free-text answer instead. The session runs right in the task chat, so everything you decide is part of the conversation the build turn reads.

The questions are grounded, not generic. If the AI asks about a validation rule or a trigger, it found that component in your org's synced metadata and will name it.

Starting a session

Type /planning in the composer, or press / to see every skill your workspace has (Planning is one entry in that list). You can also arm the composer's Plan chip before sending — on the dashboard it starts off, so a new task is an ordinary task unless you arm it first; inside a task's chat, arm it, then send your message as usual. Either way, a start card opens the session right in the transcript, and every turn from there on — starting with the message that opened it — carries the slate-colored left border that marks the session lane.

A planning session's start card in a task's chat transcript, with the triggering message directly below it inside the bordered session lane
A planning session's start card in a task's chat transcript, with the triggering message directly below it inside the bordered session lane
The start card opens the session; the message that triggered it sits inside the bordered lane from the very first turn.

One grounded question per turn

Every question names the specific thing it's worried about — a relationship type, a profile, a risk of data skew — and ends with a short numbered list of answers, the AI's own recommendation listed first. An answer it couldn't verify against your org's metadata this session is marked unverified rather than stated as fact.

A planning session's first question in the task chat, ending with a numbered list of answers and a recommendation
A planning session's first question in the task chat, ending with a numbered list of answers and a recommendation
The session's opening question: grounded in a real schema fact, with the recommended answer listed first.

Answering — a digit or your own words

Reply with the number of the answer you want, or type whatever you'd actually say — both work, and nothing blocks on picking from the list. When you reply with just a digit, the AI's next message opens by restating what you chose in plain words before moving to the next risk area, so the conversation reads back cleanly even if you only ever typed numbers.

A one-digit reply in the task chat, followed by the AI's next message restating the chosen answer in words
A one-digit reply in the task chat, followed by the AI's next message restating the chosen answer in words
A bare "1" reply, restated in words at the top of the next question — never left as just a digit in the record.

The session lane

Every turn inside an open session carries a slate-colored left border, so scrolling back through a long task you can see at a glance which stretch of the conversation was the planning session and which was ordinary work.

Several turns in the task chat with a slate-colored left border marking them as part of the planning session
Several turns in the task chat with a slate-colored left border marking them as part of the planning session
The slate border runs down the left edge of every turn between the start and end cards.

The wrap-up

A session ends two ways: the AI wraps up on its own once the major risks are covered, or you click the lit Planning chip next to the composer at any point. Either way, an end card closes the session with a plain-language summary — the decisions you confirmed and the risks you chose to accept, each named specifically and marked confirmed or flagged when the AI couldn't re-verify it. Nothing to export or re-paste: when you then ask OrgBolt to build the change, the build turn reads the wrap-up straight from the conversation.

The planning session's end card, listing confirmed decisions and accepted risks
The planning session's end card, listing confirmed decisions and accepted risks
The end card: every decision and risk the session surfaced, each marked confirmed or flagged if it couldn't be re-verified.

Building from the plan

Once the session has ended, just ask OrgBolt to build the change — a message like "go ahead and build this now based on the plan." The build turn reads the end card's decisions and risks straight out of the conversation and stages the artifacts the plan actually named, no restating required.

The task's build artifacts sidebar after a build turn that followed a planning session's decisions
The task's build artifacts sidebar after a build turn that followed a planning session's decisions
The build turn's staged artifacts, following directly from the plan's decisions — ready to review before deploying.

Before you deploy what a planning session helped shape, read /help/review — a second, one-shot Skill that audits the staged artifacts against the plan one more time.

What it costs

Planning turns are ordinary turns: they debit credits by what they actually cost, shown per turn in the chat. A typical session runs a handful of short turns. There's no separate charge for starting or ending a session.