AI AgentImplementation & QA

Discover Iris

CRO cycle · 04 · Implémentation

Stop depending on your A/B testing solution: capitalize on your own LLM and on Welyft's expertise. Iris codes the variants on your platform and QA-tests them before exposing a single visitor.

Codes the variants according to your A/B testing tool
QA-tests on desktop, tablet and mobile
Checks consent and tracking plan
Delivers the GO / NO GO QA report

Book a demo See the whole fleet

Iris

Implémentation & QA

Driven by a Welyft CRO expert

Real-world use cases

What Iris can do for you

Scenario 1 · Variant coded
Iris
Live
You : Code variant B on Kameleoon from the Drive mockup.
09:41 · Seen
IrisVariant B is coded according to Kameleoon conventions: element waiting, anti-flicker, cleanup and revert planned. It's ready for QA, with the preview URL.
Scenario 2 · QA before UAT
Iris
Live
You : QA-test variant B on AB Tasty before client UAT.
11:08 · Seen
IrisVariant checked on desktop, tablet and mobile, as new and returning visitor, before and after consent. The tracking plan hits come through; two display discrepancies are listed with screenshots.
Scenario 3 · GO / NO GO report
Iris
Live
You : Create the GO / NO GO report for this test based on the Drive test plan.
14:26 · Seen
IrisThe report covers every point of the test plan with its status and evidence. QA is done half by AI, half by a consultant: the final GO decision belongs to them.

Illustrative examples: each mission starts from your data and your tools.

Deployment

Agents trained like experts, updated like products.

Each agent is a versioned plugin on GitHub. Three ways to make the most of it.

During a mission

During the engagement

Our consultants call on the agents at every stage of the CRO cycle. You receive the deliverables, without installing anything.

Via the Welyft platform

Remote, at Welyft

The fleet runs on our platform, connected to your tools with your consent. An expert validates every deliverable.

Internalized within your organization

ConstructionWelyft builds the agent using its method, test learnings and CRO best practices.

PersonalizationEach agent is tailored to your method, brand guidelines and vocabulary.

GitHub MarketplaceYour custom agents install as plugins from a repository dedicated to your organization.

UpdatesEvery Welyft agent update is versioned, reviewed and delivered to you.

Your toolsConnected to your stack via MCP.

GuardrailsA fixed permissions matrix that decides and validates.

Within your organization

We build your agents using our method, customize them, then a GitHub plugin marketplace delivers their updates.

Let's talk about your environment Train your teams: AI for CRO

Internalize vibe coding with Iris and no longer depend on anyone to code your tests.

Iris, Welyft's implementation & QA agent, codes your variants on your A/B testing tool and QA-tests them on every device before exposing a single visitor.

Book a demo

Its skills

Code, QA, launch

Skills · Coding

Live
Based on your platform
AB Tasty, Kameleoon, Dynamic Yield, Optimizely Web: Iris codes using each tool's conventions.
/iris:coding
Live
Clean and reversible code
Cleanup and revert, anti-flicker, SPA sites, goals and transactions are handled from the start.
/iris:coding
Live
Debugging an existing test
The test doesn't display, it flickers, the goal isn't tracked: Iris diagnoses and fixes it.
/iris:coding

Skills · QA testing

Live
By device
Each variant is checked on desktop, tablet and mobile.
/iris:qa-testing
Live
Visitor and consent
New or returning visitor, before and after the consent choice: the test behaves as expected.
/iris:qa-testing
Live
Verified tracking plan
Tracking plan events are recorded for each variant.
/iris:qa-testing

Skills · Go-live

Live
GO / NO GO report
A QA report, point by point, before UAT and launch.
/iris:go-live
Live
50% AI, 50% human QA
Iris runs the checks, a Welyft consultant reviews and signs off.
/iris:go-live
Live
From your Drive
Brief, test plan and mockups are taken directly from your Drive.
/iris:go-live

Human first

Iris prepares, codes and checks, the Welyft expert validates.

Deliverable: Variantes codées et PV de recette GO / NO GO
The expert decides: le go de lancement

Its safeguards

50% AI, 50% human QA: no launch without a GO report.
Each variant is checked by device, new or returning visitor, consent and tracking plan.

Under the hood

What Iris relies on

Trained on

Plus de 500 A/B tests développés par des humains, la documentation des outils d'A/B test et les bonnes pratiques de performance web et d'accessibilité.

Data used

Brief, test plan and mockups (Drive)
Variant preview URL
Client's A/B testing tool, to force variants

Specialties

/iris:welyft-ab-coding
/iris:welyft-ab-qa

It's live : /iris:welyft-ab-coding, /iris:welyft-ab-qa

Our offers

Offers where Iris is involved

FAQ

Frequently asked questions about Iris

Une question sur votre contexte ? Nos experts vous répondent.

Posez votre question

Which A/B testing tools does Iris support?

AB Tasty, Kameleoon, Dynamic Yield and Optimizely Web for coding. The build also covers Adobe Target and VWO.

Does a human review Iris's work?

Yes. The build is 50% done by AI and 50% by a Welyft consultant, and no test is launched without a GO report.

Can Iris be entrusted with a test that's already live?

Yes: debugging a test that isn't displaying, a flicker effect, or a goal that isn't reporting, and porting a test from one tool to another.

Who gives the green light to launch?

The Welyft expert. Iris prepares the variants and the report; the launch go-ahead is theirs.

Do agents have access to our data?

Only to the sources you open, with a scope and access defined by contract. Agents work on your tools and your exports; no personal data is copied into a deliverable.

In a few seconds

Iris, from data to deliverable

An acceptance report: GO or NO GO

Implementation & QA

GO / NO GO before UAT

Variantes codées et PV de recette GO / NO GO

Variante codée, recettée sur 3 écrans.

Recipe PV

Display by device
New / known
Consent
Goal tracking

coded tools, including AB Tasty

tools tested and approved, including VWO

50% AI, 50% human

The agent and the consultant share the recipe

Anti-flicker, SPA, goals

Client-side pitfalls handled by default

Variation code, JS and CSS

GDPR and tracking verified

Illustrative demonstration: the examples are not client results.