Validation Loops
for Agentic Engineering

@ Enginy Engineering

Halftone shape

Agenda

1

Why the SDLC compressed

Validation vs verification in an AI-first workflow

2

Two live demos

Bug → regression guard and feature → validation loop

3

Rollout at Enginy

CI gates, daily suites, and expansion roadmap

Context

The SDLC didn't speed up.
It compressed.

With coding agents, handoffs merge into one fast loop. Quality must shift from late-stage gating to continuous validation.

Old bottleneck: writing code.
New bottleneck: validation speed + context quality.
SDLC compression diagram

Core distinction

Verification vs Validation

Verification Validation
QUESTION Are we building it right? Are we building the right product?
FOCUS Conformance to spec Real user outcomes
AI IMPACT Agents commoditize this Main engineering leverage point
Takeaway: Agents make verification cheap. Humans own validation.

Source synthesis

Three perspectives, one conclusion

"Software validation becomes the main engineering task."

pleasedontdeploy.com

"There are no steps. There is intent, context, and iteration."

boristane.com

"Engineers make sure agents can do the work well."

Michael Bloch
Our response: automated, tagged, evidence-rich validation loops.

qa-use + Desplega

Agent-first validation engine paired with an orchestration platform.

qa-use is the hands-on workflow. Desplega is the control plane: discovery, suites, history, and scalable execution.

Browser Automation

Agent-guided headful and headless browser runs

YAML Tests

Declarative test definitions in e2e/*.yaml

Rich Artifacts

Video, logs, and screenshots as review evidence

Suite Orchestration

Tags, schedules, CI triggers, and alerts

01

Demo 1
Bug to Regression Guard

Demo 1

Reproduce. Fix.
Prevent.

1

Reproduce

Run repro on prod with existing QA login test

2

Record

Save as e2e/bug-*.yaml with regression + bug tags

3

Verify

Same steps pass on local after fix; suite catches regressions

name: Bug repro - composer freeze
app_config: engage-web
depends_on: login
tags:
  - regression
  - bug/ENG-742
steps:
  - action: goto
    url: /messages
  - action: click
    target: compose button
  - action: to_be_visible
    target: compose modal

02

Demo 2
New Feature Validation

Demo 2

Verify first.
Record second.

Explore broadly with /qa-use:verify. Watch the agent headful. When behavior is accepted, persist with /qa-use:record.

VERIFY RECORD ENFORCE
Verify then record flow

Integration

PR Gate + Daily Suite

- name: Run regression suite
  env:
    QA_USE_API_KEY: ${{ secrets.QA_USE_API_KEY }}
  run: |
    npm install -g @desplega.ai/qa-use
    qa-use test run --all --download
  if: always()
  uses: actions/upload-artifact@v4
  with:
    name: e2e-results
    path: /tmp/qa-use/downloads/
regression gate daily suites shared alerts artifact reviews

Execution Plan

🛡

Week 1: Enforce

  • Regression suite in CI on every PR
  • Daily suites via Desplega
  • Bug fixes require a repro test

📋

Week 2: Evidence

  • Artifacts attached to PRs
  • Suites configured by tags
  • Failures alert shared channel

🔭

Weeks 3-4: Expand

  • Auto-discovery for missed journeys
  • API testing for non-UI paths
  • Reporting for flaky hotspots

Keep the speed.
Add the validation memory.

1 bug → 1 regression test Feature PRs with artifacts
Halftone stars

Questions?