AI agents · automation · AI products

AI agents that do the work, inside the teams that need them.

I design and build agents and automations for product, marketing and operations teams, then stay until they are part of the routine. QA, content, e-commerce, motion design: if a team repeats it every week, it can run itself.

See the work Start a project
agents · overviewpreview
  1. ●qa-agent run on build 3.4.1, iPhone 15 and Pixel 812 flows
  2. ✓qa-agent found 2 bugs, recordings attached4 min
  3. ●commerce-agent syncing stock and prices across 5 channels1,240 SKUs
  4. ✓commerce-agent ERP check: income, expenses, store stockclean
  5. ●motion-agent rebuilding a competitor ad with new hook5 scenes
  6. ✓trend-agent rising format spotted, alert sent to teamnow
Built for
  • Consumer app studios
  • E-commerce and retail
  • Marketing and content teams
  • Clients in Europe, the US and Dubai

Selected work

Each of these started as a person doing the same task by hand. Client and product names are kept confidential.

01 · Quality

QA agent for real devices

One prompt, and the agent plays your app on a real iPhone, catches what breaks and writes the bug report.

  • Runs iOS and Android apps and Unity games on physical devices
  • Walks through flows, spots broken screens and records a video or screenshot at the moment of failure
  • Writes a QA report with checks passed, failures and findings, then ships as a plugin each developer runs locally
Claude CodeXcodeadbUnity
qa-agent · build 3.4.1preview
  1. ●connect iPhone 15, Pixel 82 devices
  2. ●install build 3.4.1ok
  3. ●run onboarding flow6 screens
  4. ✓paywall renders, prices match storepass
  5. !bug: settings button overlaps on Pixel 8rec_014.mp4
  6. ✓report written with 2 recordingsdone

02 · Operations

E-commerce operations agent

A jewelry brand's catalog, stock and prices stay in sync across every sales channel without anyone copying data by hand.

  • Publishes new products to e-commerce sites and marketplaces
  • Keeps stock and prices aligned across channels and stores
  • Runs regular income, expense and store stock checks on ERP data and flags gaps
PythonLLM APIsNebim ERPMarketplace APIs
commerce-agent · nightly runpreview
  1. ●pull new products from ERP38 items
  2. ●write listings and publish3 channels
  3. ●compare stock across stores12 stores
  4. !price mismatch on 4 items, fixedsynced
  5. ✓income and expense checkno gaps
  6. ✓summary sent to operationsdone

03 · Creative

Motion design agent

Describe the video in one prompt and get a finished, ready-to-post cut. No manual edits.

  • Plans the edit from a single prompt and confirms open questions before it starts
  • Cuts the footage, scores music to the beat, adds transitions, SFX and on-screen copy
  • Delivers reels and app ads in every format, with quality checks and memory of past projects
Claude CodeRemotionModel router APIDashboard
motion-agent · run 0412preview
  1. ●analyze competitor_ad.mp4hook 0-2s
  2. ●storyboard with new hook5 scenes
  3. ●generate scenes5/5
  4. ●voiceover and captionsdone
  5. ✓QC: text fits, brand colorspass
  6. ✓export 9:16 · 1:1 · 16:9for review
ugc-workflowpreview
  1. ●brief from product page
  2. ●script 3 creator angles
  3. ●produce videos
  4. ✓schedule to channels

Content

UGC creator workflows

AI presenter reels with a hook, on-screen copy and a call to action, produced and scheduled for brands' social channels.

Workflow: a daily schedule reads an account list, scrapes each Instagram account, keeps today's posts and sends new ones to Slack.

Social

Social media trend agent

Checks a list of Instagram accounts every day, keeps only the new posts and alerts the team in Slack, so trends are spotted the day they start.

screenshot-agentpreview
  1. ●read brand tokens
  2. ●build static and dynamic sets
  3. ✓fit and contrast checks
  4. ✓place in Figma, localize

App Store

Screenshot agent

Builds static and dynamic App Store screenshot sets from brand tokens, as editable layers in Figma.

Under the hood

Workflows behind the work

Workflow: at 09:00 and 18:00 it reads fal.ai and EachLabs balances and usage, posts a summary to Slack and sends direct messages to a group when spend passes a limit.
AI spend alarm. Reads image and video API balances and usage at 09:00 and 18:00, posts a summary to Slack and messages the owners directly when spend passes a set limit.
A large node graph that fans one prompt out into 25 parallel image generations and collects them into one output.
Batch image pipeline. One graph that fans a single brief out into 25 colored manga images and collects them in one run, ready for layout.

What I build

Work that ends with something running in production, used by the people it was built for.

Agents for internal work

QA, reporting, research and the repetitive steps between tools. Built on Claude and Claude Code, packaged so teams run them without me.

Creative and content automation

Ad videos, UGC, store screenshots and social content produced by agents and reviewed by people, using image, video and voice models.

Operations and integrations

Connecting AI to the systems a business already runs on: ERP, e-commerce platforms, marketplaces, Slack and internal databases.

AI product strategy

Which features to build with AI, which model fits each one, what it costs per run and how to price it. Learned leading an AI product studio.

How a project runs

Short loops, real users from the first week, and a handover at the end.

i.

Find the stuck step

Interviews and time with the people doing the work, to find the task that costs the most each week.

Week 1
ii.

First working version

A prototype on the tools the team already uses, tested on real tasks instead of demos.

Weeks 1 to 2
iii.

Ship and hand over

Packaged, documented and trained, so the team runs it without me.

Weeks 3 to 4
iv.

Measure and extend

Track the time, cost or quality it changed, then pick the next workflow.

Ongoing

I spent eight years as a full-stack engineer before moving into product. Today I lead an AI product studio and still build the agents myself, so I can talk strategy with a founder in the morning and debug an integration in the afternoon.

12years building software
13person studio led today
4AI apps shipped in 6 months
100K+downloads across those apps

Daily tools: Claude, Claude Code, OpenAI, Gemini, n8n, fal.ai, ComfyUI, ElevenLabs, Remotion, Python, FastAPI, React, Next.js, SwiftUI, Figma.

Contact

Have a workflow that should run itself?

LinkedIn