Slingshot does the work behind every deal. You say go.
An agent on Astro for account executives and sales engineers. It investigates the account, cites what it finds, drafts the deck, the plan or the message, and waits for your Go / No-Go.
Why it is here
Discovery notes are logged on the deal. No playback deck exists yet. Flagged by a fixed rule, not a model.
opportunity:opp_106
Pass 1
Investigate. Gathered what was said in discovery and what changed on the deal.
Judge. Sent back. One claim has no source.
Pass 2
Investigate. Found the call the claim came from and cited it.
Judge. Through. Every claim cites its evidence.
Draft
Writing. The playback deck, with a demo path.
Draft, waiting for you
Discovery playback deck, with a demo path. Two passes, every claim cited.
call:disc_0412opportunity:opp_106notes:fennimore_03
Acted on your Go
Deck saved to the deal and shared with the account team. Simulated: nothing leaves the demo.
Approved by ae.demo
The work that slips between calls
Every deal and account throws off the same chores, and they wait for a gap between calls. Slingshot spots the gap from fixed rules and does the first full draft, so what reaches you is something to approve, steer or dismiss.
16 plays in the library. Five of them:
| When the board says | Slingshot drafts | You get |
|---|---|---|
| Discovery notes, no playback deck | Discovery playback | A deck that plays back what the customer told you, with a demo path built on it. |
| No PoC success criteria yet | Proof-of-concept plan | Success criteria you can measure, and the plan that proves each one. |
| An account graded at risk | Risk outreach | The message to send, with the reasons behind the grade beside it. |
| A call with no follow-up sent | Follow-up | The follow-up after the call, written from what was said on it. |
| No next step logged | CRM update | The changes laid out for approval before anything is saved. |
Two loops, and you hold the exit
A play circles until the work is right. The first loop is between the agent and a judge. The second is between the draft and you. Nothing gets out of either without clearance.
- It investigates, and cites
- It reads what was said and what changed, and attaches the evidence to each claim. A judge rejects any claim that arrives without a source, and the pass goes round again.
- It drafts the real thing
- Not a summary of what a deck could say. The deck, the plan, the message, written for an account executive or a sales engineer.
- It waits for your Go / No-Go
- You approve, steer or dismiss, with the evidence beside the draft. Steer it and your feedback becomes the brief for the next pass.
- It acts only on approval
- Sending, saving and sharing each wait for a person. A No-Go ends the play with nothing sent.
What always holds
-
Every claim cites its evidence.
Open a strip and the source is one click away. An unsupported sentence does not reach your draft, because the judge has already sent it back.
-
Grades come from fixed rules, never from a model.
The board shows which rules fired and the points each one carried. The same deal gets the same grade every time.
-
Nothing is sent without a person approving it.
The agent can draft all day. Sending, saving and sharing are yours to release, one approval at a time.
-
Each person sees their own deals.
Your board holds your deals and accounts, and the plays drafted for you. Nobody reviews a draft that is not theirs.
Airflow 3.4 on Astro runs every play
A play is an Airflow run. You never have to open it, and you always can: the run behind any play is one link away, with every pass, review and approval in it.
- Task Loops
- The investigate and judge cycle and the draft and review cycle are loops inside one run.
- Agent tasks
- Agent tasks reach their models through the Astro LLM gateway.
- Human review and tool approval
- Your Go / No-Go and each approval happen inside the run, which waits for you.
- Durable retries
- A step that fails is retried without losing the work that came before it.
- Event-driven triggers
- A change on a deal or an account can start a play.
- History replay
- The demo's history is replayed through the engine from start to finish.
- 180 days of historyreplayed in under 4 min
- The LangGraph rebuild
- The same engine is rebuilt in LangGraph and the two are measured side by side. The scorecard is honest about both.
One strip is still waiting on you.
Open the board, start a play on a demo deal, and give it your Go or your No-Go.