BETA
Get started 
Think and prove at the pace you build.
AI has solved the problem of how fast developers can build. The urgent problem now is for teams to think about what is worth building and prove that it works at the same pace.
The control center for a digital product led business. One canvas where your team and AI operators research, analyze, design, build, test, monitor and improve, together.
By Scopeout
Control center for digital product led businesses
A business that runs itself. With your team.
Scroll to read↓The problem
AI made us faster at shipping. It didn't make us faster at knowing what to ship for better results.
01
The arc
From weeks of guessing, to a single sentence, to products that run themselves.
Three eras of building products, each shorter than the last. AI has solved how fast developers can build; the urgent problem now is for teams to think about what is worth building and prove that it works at the same pace. Scopeout is the bridge from the second era to the third.
01
Then2010 to 2023
02
Now2026
03
Next (Scopeout beta)Live today
Backlog 14
Discovery 3
v7 · v5 · v3?
Design 2
unlinked
Dev 1
caused drop
S#product
Ppriya
Ttejas
decision lost6 wks
Weeks of guessing.
Research in one tool, designs in another, tickets in a third, numbers in a fourth, the decision in a chat thread. Six weeks from question to fix, and nobody checks if it worked.
Prompt
Add saved addresses to mobile checkout.
AI coding agent
Delivery address
HomeUse
OfficeUse
Continue
PRTestsPreviewFlag
shipped0:42
✕trace✕review✕team
Why this?unknown
A sentence becomes a feature.
One prompt: code, tests, preview, live in an afternoon. But which sentence to type is still a guess, made alone, with the reasoning lost.
Originbaseline
P
−6.2%Loop 1APS+1.8% lift
P
strategyresearch
A
designS
buildP
testlearn
Loop 2ADPS+3.9% lift
P
strategyP
researchA
designS
buildtest
D
learnLoop 3ADPS+7.2% lift
D
strategyresearch
A
designS
buildP
testD
learntracedPapproved✓promoted
This is Scopeout.
Scopeout continuously optimizes your product, spinning up new onboarding, pricing, flow, layout, and copy variants and running them as live A/B tests against real traffic. Winners get promoted, losers get retired, and the next round of variants is already in flight. Your product gets sharper every week, on its own.
02
What the control center runs
Seven buckets. One canvas.
The loop every product team already runs, made operable: Research, Analyze, Design, Build, Test, Monitor, Improve. Pick a bucket. See the prompt, what the canvas ran, and the artifact it produced, plus the teammate who directed it. Screens marked "real run" are unedited captures from Scopeout on 14 Sep 2026.
It gathers your evidence
Ask a question. Get a plan, then evidence.
Type the problem. The canvas proposes the steps, asks for data it does not have, and stops for your go-ahead before spending a token. Then Research reads replays, tickets, reviews, transcripts and competitor flows, and returns a cited brief.
On the canvasResearch runs it · PM directs · anyone can ask
Real run · plan + permission gateIt turns evidence into decisions
Findings, fixes and priorities, ranked on the board.
Analyze consolidates what Research found into boards: findings in blue, fixes in green, priorities in yellow, plus a comparison table. Hypotheses are ranked by impact, confidence and effort. The Research Agent stops for approval before anything is designed.
On the canvasPM directs · Design and Eng review · leadership sees the why
Real run · consolidated findingsIt designs in your system
On-system screens, not generic ones.
Give it the spec, a screenshot or a sentence. Design builds the screen from your imported design system: here, a delivery address step with progress, Apple Pay shortcut, saved addresses first and a single-line address field. The designer picks and refines inline.
On the canvasDesigner directs · PM comments · Eng gets the build-ready variant
Real run · generated from the design systemIt ships in your codebase
From chosen variant to pull request.
Connected to your repo and design system, Build turns the chosen variant into working code behind a flag, writes the tests, deploys a preview and opens the PR. Engineers review code that already matches the spec and the evidence.
On the canvasEng directs and reviews · Design signs off · PM tracks against the spec
Prompt · Sam, Eng
Implement variant B behind a flag and open the PR.PR #2231 · single address field
feat(checkout) · +214 −88 · preview live
AddressStep.tsx +131 −62 useAddressLookup.ts +58 checkout.spec.ts +25 −26 tests: 12 pass flag: address_v2
Linked · H1 · Brief · PRD · Test4 refs
Review · SamMerge
It proves the change
A/B/n with gates, not vibes.
Test runs the variants, computes lift with sequential stopping rules, holds guardrails on refunds, support and latency, and promotes the winner only when it clears the bar you set. Every result is written back as evidence for the next run.
On the canvasGrowth directs · Eng approves the flag · leadership reads the result
Prompt · Priya, PM
Start a 7-day test of variant B with guardrails.Test · address-step-v2
Day 3 of 7 · 21,804 sessions · sequential
Confidence93.8%
Guardrailsholding
Auto-promote at 95% or moreRunning
It watches your product
Find the leak before the dashboard does.
Every funnel step, segment and release compared against its own baseline. When a metric moves, Monitor isolates where, for whom and since when, then queues the problem for Research with an owner, instead of waiting for Monday.
On the canvasGrowth directs · runs on its own · PM gets the queue
Alert · monitor
Mobile checkout −6.2% at address entry since 4.12.0.Funnel · checkout, mobile, 7 days
iOS Safari only · desktop unchanged
Cart100%
Address58% · −6.2
Payment49%
Queued · Research · owner PriyaInvestigate
It makes the next run smarter
Every decision recorded. Every outcome learned.
Each run writes its own record: problem, evidence, decision, who approved, outcome. Results feed the outcome library and your playbooks, so the next hypothesis starts from what worked. Leadership reads the log; nobody writes the update.
On the canvasLeadership reads · every team writes, without writing
Prompt · Dana, COO
What shipped this week, and did it work?Growth review · week 37
3 runs · 2 shipped · 1 running
Problem · mobile checkout −6.2%run #418
Evidence · 12 sourceslinked
Approved · Priya, Arjun, Sam3
Outcome · +42% step conversionverified
Auto-writtenShare
03
Anatomy of a run
One sentence. Six operators. Three teammates. A real change.
Steps 1 to 3 are what Scopeout actually did on 14 Sep when we typed this sentence; the capture is unedited. Steps 4 to 6 are the intended flow once Build and Test are wired to your repo and flags.
“Mobile checkout conversion dropped 6% this week. Find out why and fix it.”
01
detect-anomaly · 00:00:41
Drop isolated to one step, one segment
Address entry, iOS Safari, since release 4.12.0. Desktop and Android unaffected.
02
gather-evidence · 00:11:08
Replays, tickets, reviews, a live walkthrough
142 sessions, 38 tickets, 27 reviews; the browser agent fails autofill 4 of 5 times on the real checkout.
03
synthesize-hypotheses · 00:14:22
Findings, fixes, priorities on the board
Boards with blue findings, green fixes and yellow priorities, a comparison table, and ranked hypotheses. The Research Agent stops: "Diagnosis findings and prioritized hypotheses must be reviewed and approved before investing in the redesign."
Priya (PM) approves the diagnosis · 00:19
04
generate-variant · 00:31:05
Three on-system variants
Built from your components; frames sized to content.
Arjun (Design) picks B, edits inline · 00:44
05
open-pull-request · 00:58:47
Code behind a flag, tests, preview
PR #2231, +214 −88, 12 tests pass, preview deployed on address_v2.
Sam (Eng) reviews and merges · 01:52
06
start-test → log · 02:13:40
Seven-day test, written to the record
Sequential A/B with guardrails; the whole run logged for the week-37 growth review.
40 minutes in (real run)
address-step-v2Running · day 3 of 7
No research deck. No spreadsheet export. No status meeting. The growth lead typed one sentence, three teammates approved in place, and the record wrote itself.
Open this run on the canvas→04
The same canvas, in seven buckets at once
A complete product and business platform, run by one canvas.
Each bucket below is a real area of the canvas. Buckets marked "real run" show unedited captures from 14 Sep 2026; "concept" marks what is designed but not yet wired. Nothing ships as a separate tool your team has to adopt, and every bucket has a person who directs it.
Real run · 14 Sep 202601 / 07Directed by PM · anyone can ask
Research
Evidence, not opinions.
Ask a question in a sentence or drop a screenshot. The canvas plans the work, asks for the data it does not have, and waits for your go-ahead. Then replays, tickets, reviews, interviews and competitor teardowns become a cited brief on the canvas.
In this bucket:
Open the Research bucket→Plan first, permission before spendingorchestrator
Replays, tickets, reviews, transcriptssource connectors
Live product and competitor walkthroughsbrowser agent
Real run · 14 Sep 202602 / 07Directed by PM · Design and Eng review
Analyze
Evidence becomes a decision.
Findings, fixes and priorities land on boards the whole team can see. Hypotheses are ranked by impact, confidence and effort; the spec and acceptance criteria link back to the evidence. Approval happens on the block, not in a meeting.
In this bucket:
Open the Analyze bucket→Comparison boards and tablescanvas blocks
Hypothesis rankingimpact × confidence × effort
PRD and acceptance criteriaJira · Linear
Real run · 14 Sep 202603 / 07Directed by Designer · PM comments
Design
On-system, sized to content.
Generate flows and screens from your design system: Figma, Storybook or code. Give it a screenshot and get on-system variants, side by side. Edit inline; the designer keeps the pen.
In this bucket:
Open the Design bucket→Screenshot or spec to on-system UIA2UI
Variants, sized to contentno scroll inside frames
Design-system importFigma · Storybook · live site
Build · concept
feat(checkout) · single address field+214 −88
AddressStep.tsx +131 −62 useAddressLookup.ts +58 checkout.spec.ts +25 −26 tests 12 pass · preview address_v2
SMatches the spec. Merging.
04 / 07Directed by Engineering · Design signs off
Build
Your codebase, your flags.
The canvas reads and edits your actual codebase. Prototype, tests, preview, pull request, behind a flag and linked to the spec and the evidence. Engineers review; nothing ships without them.
In this bucket:
Open the Build bucket→Prototype to pull requestGitHub · GitLab
Tests, preview deploy, verificationsandbox
Feature flagsLaunchDarkly · custom
Test · concept
address-step-v2day 3 of 7
Confidence · sequential93.8%
Guardrailsholding
05 / 07Directed by Growth · Eng approves the flag
Test
Gates, not vibes.
A/B/n tests with sequential stopping rules and guardrail metrics. Winners promote automatically once they clear the bar you set; results write back to the canvas as evidence for the next run.
In this bucket:
Open the Test bucket→A/B/n with sequential stoppingstats engine
Guardrails: refunds, support, latencymonitors
Auto-promote behind approvalflags
Monitor · concept
checkout · mobile · 7dflagged
Cart100%
Address58%
Payment49%
Since release 4.12.0iOS Safari
06 / 07Directed by Growth · runs on its own
Monitor
Find the leak first.
Every funnel step, segment and release watched against its own baseline, deterministically, around the clock. When something moves, Monitor isolates where, for whom and since when, then queues the problem for Research with an owner.
In this bucket:
Open the Monitor bucket→Funnel and retention monitorsPostHog · GA4 · Mixpanel
Anomaly and release correlationbaselines
Queue with an ownerSlack · Teams
Improve · concept
growth review · week 37auto-written
Problem · checkout −6.2%run #418
Approved · Priya · Arjun · Sam3
Outcome · +42% step conv.verified
DThis is the whole review. Nice.
07 / 07Leadership reads · everyone writes, without writing
Improve
The record writes itself.
Every run writes its own record: problem, evidence, decision, who approved, outcome. Outcomes feed the playbooks, so the next hypothesis starts from what worked. Leadership reads the log in Notion, Confluence or Slack; nobody writes the update.
In this bucket:
Open the Improve bucket→Decision log and outcome libraryfirst-class record
Playbooks and proven methodsskills
Weekly digestNotion · Confluence · Slack
05
Collaborative by default
Operators do the running. People do the directing.
One canvas, the whole company on it. Product, Design, Growth, CX, Engineering, and leadership work on the same blocks, see the same evidence, and approve in place. No exports, no status decks.
01 · Memory
Shared context, once.
Your product, customers, brand, design system, codebase, and past decisions live on the canvas a single time. Every operator and every teammate works from the same memory.
product · customers · brand · code · decisions
02 · Control
Approvals per part.
Decide who directs each part: the PM approves hypotheses, the designer picks variants, the engineer merges, growth promotes winners. Autonomy is a dial, per part, per team.
pm · design · eng · growth · leadership
03 · Flow
Hand-offs that carry context.
When Research hands to Product, or Design hands to Build, the evidence and the reasoning travel with the block. Comments, mentions, live cursors. Nothing is re-explained in a meeting.
mention a teammate or an operator the same way
06
What's next
An autonomous growth engine, running on your product.
Once a product has traction, the same canvas that built the fix starts looking for what to improve, on its own. It finds the growth leaks, writes the hypothesis, runs the test, ships the winner. You watch the numbers move.
01Detection
A continuous engine watching every funnel.
Funnels, anomalies, friction scoring, segment discovery, release correlation. All deterministic, all running 24 / 7. Causal-inference and significance guards built in so the canvas only acts on real leaks, not noise.
funnel monitorsdetectorsrollups
02Synthesis
Signals become hypotheses you can ship.
The engine produces a ranked queue of structured signals. The model sits at the very end as a synthesizer, pulling replays, reviews, tickets, and your team's notes, and writes a human-readable hypothesis with proposed test, expected lift, and confidence.
structured signalsevidence raghypothesis writer
03Learning
Every test makes the next one smarter.
A growing library of your past outcomes, proven methods, and your team's expertise feeds the synthesizer, so winning patterns on one part of the product accelerate the next. Benchmarks against products like yours show where you stand.
outcome libraryplaybooksbenchmarks
04Ship the winners
A / B / n today. Bandits tomorrow.
Tests start on the parts that move outcomes most: onboarding, activation, pricing, checkout, then flows and copy. Proper A / B / n with significance gates and guardrails; winners promote automatically, and your team decides where the autonomy threshold sits.
onboardingpricinga/b/n → bandits
Available for products with traction. Onboarding is hand-picked while the engine is in beta; the canvas, research, design, and build parts are open to everyone.
Contact us about the engine→07
Why now · who it's for
Speed moved. The bottleneck moved with it.
Why now
Who it's for
Teams that own outcomes, not tickets.
Built for Product, Design, Growth, and CX teams, and the leaders who run the business, at startups and enterprises that need to identify customer experience problems faster, fix what matters, and drive business outcomes with documented visibility throughout.
ProductDesignGrowthCXEngineeringLeadership
08
FAQ
Questions, answered.
How is "agentic" different from a chatbot?
A chatbot answers a question and forgets it. Scopeout's operators act on the canvas with shared context (product, evidence, design system, past outcomes) and hand work to each other: Growth detects, Research gathers, Product specs, Design builds, Experiments proves, Visibility records. You direct; the canvas operates.
Can the AI break my product?
No operator ships to production on its own. Build opens pull requests behind flags; Experiments promotes only past the significance and guardrail gates you set; every change is reversible from the record. Autonomy is a dial per part. Turn it up as trust grows.
How does my team work together on it?
Everyone is on the same canvas. A growth lead types the problem, Research gathers the evidence, the PM approves a hypothesis in place, the designer picks a variant, the engineer merges the PR, and leadership reads the record, on the same blocks, with comments and mentions, each hand-off carrying its context. Roles decide who directs each part.
Do I have to use the AI for everything?
No. The canvas works as a shared workspace for product context, research, specs, and designs on its own. Each part is a switch: run Research by hand, let Growth watch on its own, keep Build off until your engineers want it.
What does it connect to?
Analytics (GA4, Mixpanel, Amplitude), support (Zendesk, Intercom), design (Figma, Storybook), code (GitHub, GitLab), docs and chat (Notion, Confluence, Slack), and a browser agent for anything with a URL. Connect what you have; the canvas fills gaps with evidence it gathers itself.
What about a product with years of history?
That's the point. Scopeout learns your product context from what already exists (code, docs, tickets, past experiments) so the first hypotheses come from your evidence, not a blank page. Enterprise plans add private deployment and SSO.
Stop running the business. Start directing it.
Free to try. No credit card. Cancel any time.