How the AI helps an account share a social post
This is the "Share a social post" goal from AI partner onboarding, shown as the account user sees it. Each screen is one moment in the chat. Under it, a short note says what the AI looks at, what it does, and what it shows.
Grey dashed boxes marked "For devs" are optional. You can skip them. Source: the goal spec draft (share-social-post-agent-native.md, v2).
Maya starts the goal
Maya sees "Share a social post" in her onboarding list. She taps it. She could also type the same thing into Ask AI at any time.
What happens
- 👀AI looks atWho Maya is, her company (Northwind IT), and what her plan allows.
- ⚙️AI doesStarts the same steps either way. Tapping the goal and typing the request lead to the same flow.
- 🖼️AI showsNothing yet. It moves straight to step 2.
The AI asks one quick question (only if it needs to)
The AI first checks if it already knows what Northwind IT sells. If it does, it skips this step. If not, it asks once, with tap-to-answer options.
What happens
- 👀AI looks atWhat it knows about Northwind IT. Then the solution list the vendor set up (Security, Cloud, Networking).
- ⚙️AI doesAsks one question at most. Never two before showing posts.
- 🖼️AI showsA question card with chips. If there is no list, just a text box.
get_account_interest, then list_solution_options. Inputs are what the AI reads first (here: tenant, account). Output is what comes back (the chip options).The AI shows three posts that fit
The AI picks three ready-made posts from the vendor's content library about Security. Maya can share one as is, edit it, or ask for different ones.
What happens
- 👀AI looks at"Security", the content library, and posts already shown (so it never repeats).
- ⚙️AI doesFinds the three best matching posts.
- 🖼️AI showsThree post cards with Share and Edit, plus "Show different posts".
recommend_posts. The AI only "reads" titles and ids. The card gets the full preview and images.Maya tweaks the post (optional)
If Maya taps Edit, the social post editor opens full screen. The chat stays on the left, so she can ask the AI to change the post or type changes herself.
Zero trust in 5 steps. Most breaches start with one stolen password. Northwind IT can help you close that door.
What happens
- 👀AI looks atThe chosen post and every edit Maya makes, as she makes it.
- ⚙️AI doesRewrites on request. Keeps track of the latest version.
- 🖼️AI showsThe existing editor, with chat on the left and a live preview.
open_post_editor. Edits flow back to the AI as events (a note that the user did something), so the AI never has to ask.Maya connects LinkedIn (only if she hasn't yet)
Before posting, the AI checks which social accounts are ready. If none are, it shows a connect card. Maya always starts the connection herself.
Sign in to LinkedIn and allow Northwind IT portal to post for you.
What happens
- 👀AI looks atMaya's connected networks, and which ones the vendor allows.
- ⚙️AI doesSuggests connecting. It can't connect for her. She clicks.
- 🖼️AI showsOne row per allowed network: connected, needs reconnect, or Connect.
list_social_connections, then connect_social_account (needs a user click). This also completes the "Connect a social account" goal.Maya picks when, then clicks Schedule
The AI fills in what it can (LinkedIn, tomorrow 9:00 AM). Maya checks it and clicks. Even if she types "just post it", the AI waits for her click.
What happens
- 👀AI looks atThe final post (with edits), connected networks, vendor rules, and "tomorrow morning".
- ⚙️AI doesPrefills the card. Never posts on its own.
- 🖼️AI showsA schedule card. After the click, a short confirmation with Undo.
schedule_post. Needs a user click. Safe to retry, so a double click never posts twice.Done, and here's what's next
Once a post is scheduled or posted, the goal is ticked off. The AI suggests the next goal. This also happens if Maya shares a post outside onboarding.
What happens
- 👀AI looks atThat the post went out (or is scheduled), and which goals are left.
- ⚙️AI doesMarks the goal done. If it was already done, nothing changes.
- 🖼️AI showsA done card with the next goal. If none is left, just "Done."
complete_goal. Run by the system, not the AI.One card, many places
The post card Maya saw in "Pick a post" is not a one-off. The same card, and the same connect and schedule cards, appear wherever an AI feature needs them.
When things go differently
Same steps, different paths. These are the cases to cover in mocks.
"Post about security on LinkedIn tomorrow 9am." No question. Security posts appear. Schedule card is prefilled.
The vendor hasn't set one up. The AI asks one open question with a text box, no chips.
X shows "Needs reconnect". Maya can still post to LinkedIn.
The AI shows the schedule card and waits. Nothing goes out without a click.
"Social posting isn't on your plan." No post cards.
"No posts for this solution yet." Offer other solutions.
The words in the spec, in one line each
You don't need these to design. They help when reading the full spec or talking with dev.
| Word | Plain meaning | In this example |
|---|---|---|
| Goal | Something the account wants to get done. | Share a social post |
| Action | One thing the system can do. Screens and the AI both use the same ones. | Find posts, connect LinkedIn, schedule |
| Inputs | What the AI reads before doing an action. | "Security", posts already shown |
| Output | What comes back after the action. | Three posts |
| UI block | A card that shows an output and lets the user act. Reusable across goals. | Post list card |
| Plain text | When the AI just answers in words, with no card. | "Here are 3 security posts." |
| Event | A note to the AI that the user clicked something. | "You chose: Zero trust in 5 steps." |
| Recipe | The usual order of steps for a goal. | Steps 1 to 7 in this walkthrough |
| Needs confirm | The AI can suggest it, but only the user's click makes it happen. | Connect, Schedule, Post now |
| MCP app | A way to show these same cards inside other AI tools later. | Later. V1 is still being checked with dev |