Describe a process once. Gameplan wires up every step, picks the best model for each one, and runs it on schedule. Days of setup, done in one sentence.

New system · Weekly client ad reportBalanced
Describe the work⏎ Build system

01
Every Monday, 8:00
ScheduleTrigger
02
Pull the results
—No AI needed
03
Flag anything unusual
—prompted by Claude Opus
04
Figure out why
—prompted by GPT
05
Write the summary
—prompted by GPT
06
You approve
—Only if budgets change
07
Send to the client
—App
Reading your process… Steps0Models0Per run$0.00Wired in0:00
08:00:00Trigger set · every Monday 08:00, client time zones
08:00:02Connected Google Ads and Meta · 14 accounts
08:00:09Test run · 14 clients · 0 errors
08:00:21Quality 9.0 · $0.38 per report · ready to schedule
08:00:22Approval rule · only when a budget changes over 20%

Picks from every major model, and connects the apps you already use

The problem / 01

Wiring an AI system takes days. You shop for models, write the prompts, connect the apps, test every edge case, then babysit it when something breaks. Gameplan does all of it from one prompt. Then it keeps it running.

How it works / 02

From a paragraph to a running system.

You explain it once, the way you'd brief a new hire. Gameplan does the architecture.

  1. 01Describe

    Say what you need, in plain words.

    No flowcharts, no node editors. Gameplan reads your process and picks out every action, schedule and rule in it.

    Describe

    Explain the work like you would to a new hire.

  2. 02Plan

    It splits the work into steps.

    Each step gets an impact score: how much it shapes the final result. That decides where quality is worth paying for.

    Plan

    It's split into steps, each weighted by impact.

  3. 03Match

    Every step gets its own model.

    Small models for volume, frontier models where the outcome is decided, plain code where no AI is needed.

    Match

    Each step gets the best model or app for the job.

  4. 04Tune

    Instructions, written for each model.

    Every model reads prompts differently. Each step is prompted by the model that writes best for it.

    Tune

    Each model gets instructions written the way it understands best.

  5. 05Build

    Wired up and tested on your data.

    Apps connected, checks in between, every edge case you mentioned turned into a test before anything goes live.

    Build

    Everything is wired up and tested on your data.

  6. 06Run

    It runs. You only step in when it matters.

    On schedule, every week. It asks for your approval only when a decision is genuinely yours.

    Run

    On schedule, asking you only when it should.

Most are. Gameplan asks follow-up questions about the cases you didn't mention, like a missing account or a client with no data that week. Then it shows you the full plan before anything runs, and you can change any step, model or approval rule.

Model testing / 03

We test every model. So the choice is never a guess.

Every model is scored on real business tasks, per type of task, and re-tested with every release. A model can be brilliant at extraction and average at writing. Gameplan knows the difference.

Spot anomaliesScore / 100
Illustrative scores. Highlighted is picked on Balanced.

Tests are built from real business processes, not public benchmarks. Each model is scored per type of task, because a model can be excellent at extraction and average at writing. New models and updates go through the same tests, usually within days of release. A switch is only suggested after the new model beats the current one on your own past runs.

Real work, not benchmarks

Messy data and the edge cases public leaderboards skip.

Every release, re-tested

New models usually go through the same tests within days.

Your runs decide

Nothing switches until it wins on your own past runs.

Release feed
New model released, tests runningClaude Sonnet updated, re-tested in 19 hGrok update, writing score +2Llama new version, local tests passedGemini Flash update, no changeNew model released, tests runningClaude Sonnet updated, re-tested in 19 hGrok update, writing score +2Llama new version, local tests passedGemini Flash update, no change
Prompting / 04

One model does the work. Another writes its instructions.

Every model reads instructions differently. Gameplan tests which model writes the best prompts for each one, and uses the winner on every run.Each model gets its instructions from the model that writes best for it.

4writers tested for every model
+3points on average vs. a model writing its own
Writing the instructions
Doing the workClaude Opus

Illustrative scores out of 100

Every pairing is re-tested when a model updates. And a step is re-tuned as soon as its results start to slip.

Goals / 05

Tell it what matters. It decides which model does what.

Pick one goal. Gameplan spends where the outcome is decided and saves everywhere else.

Model mix · Weekly client ad reportImpact on result
Figure out why

Where the outcome is decided

Claude Opus · prompted by GPT
Balanced
Cost per client reportPer report$0.38
Monthly, 14 clients every weekMonthly$23
Quality scoreQuality8.9

Example estimates. Real figures come from scoring your own runs.

Every step gets an impact score: how much it shapes the final result. Your goal sets how much quality is worth paying for at each level of impact. Low-impact steps go to small, local or no-AI options. The steps where the outcome is decided get frontier models. You see the cost and quality change before you commit.

Balanced isn't medium everywhere. It's cheap where nothing is at stake, and premium where the outcome is decided.

What would your work cost to run?

Move the sliders. Gameplan prices every step for the job it does.

Hours back every month77

Time your team spends on the decisions, not the copy-paste.

One frontier model on every step $114
Gameplan, Balanced $21
Saved vs. one model81%
Cost per report$0.38
Dashboard / 06

One dashboard for every system you run.

What ran, what it cost, and what needs you. The process, never the plumbing.

Built to keep running / 07

Set it up once. It keeps getting better.

Re-tested on every release

When a model ships or updates, it runs your tests within days.

Claude Sonnet re-tested19 h
Grok writing score +22 d
New model queued, 240 casesnow
Llama local tests passed3 d
Gemini Flash no change4 d
Claude Sonnet re-tested19 h
Grok writing score +22 d
New model queued, 240 casesnow
Llama local tests passed3 d
Gemini Flash no change4 d

Approvals only where it matters

You set the rules. Everything else runs without asking.

2 of 14 summaries need you

Both mention a budget change over 20%.

Approve bothReview
12 sent automaticallyRule: budget change > 20%

Fixes itself when results slip

Quality drops, the step is re-tuned.

Quality 8.4 on step 3Re-tuned · 9.1

Every app connected

Your tools, wired in once.

Gmail Slack Google Ads Sheets HubSpot Meta

Cost per run, always visible

Down 6% this month, quality steady.

$5.31per run
↓ 6% vs last month

Code you never have to read

Gameplan writes and maintains every line, every prompt, every retry. It's yours if you ever want to look.

weekly-client-report.tsWritten by Gameplan

Edge cases become tests

"What if a client has no data that week?" It asks, then checks it on every run.

Client with no spendpass
Account access revokedpass
Currency changed mid-monthpass
Duplicate campaign namesasks you
Private beta

Stop managing agents. Start running systems.

Built with a small group of teams who run complex work every week. Join the waitlist and we'll set up your first system with you.