How we work: from finding to commit
Overview
This is not a toy demo.
It is a practical look at how we actually work – applied to one of our own systems, the AutomateThis! website:
We read the site critically, collected findings, made decisions in the conversation and shipped the changes directly in the repo: committed, reviewed and deployed. An AI agent (OpenClaw) was part of the toolkit – not the point.
The goal was not simply to “generate copy.” The goal was to:
- read the website critically,
- identify positioning contradictions,
- make decisions in-context,
- and implement the changes directly in the repo.
The starting point
Over the years, the website had accumulated a familiar set of problems:
- a stronger homepage than supporting pages
- older offer/funnel language from previous phases
- legacy blog content heavily shaped by older Mautic / marketing automation framing
- footer, contact, and conversion layers still carrying SaaS/product signals
In short:
There were strong pieces, but the overall message was inconsistent.
What we actually did
The workflow looked like this:
1. Finding: read the live website
The first step was to read the public site and evaluate it through the lens of positioning, target customer, offer structure, and conversion logic.
2. Decision: clarify the real customer and offer
We then extracted the implied target customer and tightened the positioning: AutomateThis! as the engineering team that builds, connects and runs the systems behind e-commerce businesses.
3. Implementation in the real project
The work happened in the website’s actual Astro codebase. Changes were implemented for real – not just suggested.
Among other things, we:
- rewrote positioning pages
- aligned German and English offer pages
- cleaned footer and contact pages
- removed fake scarcity and stale trial/product language
- reframed legacy blog content strategically
- fixed HTML entity rendering issues centrally in the blog layout
4. Commit, review and deploy
Changes were committed continuously and pushed to main after approval.
Deployment then happened automatically via GitHub → Coolify.
Why this mattered
The interesting part was not that text got written.
It was the combination of:
- analysis
- strategy
- direct file edits
- git-based execution
- human review and steering in the same loop
That is a different operating mode from the usual brief-and-copy ping-pong.
What this way of working gets you
1. Less handoff loss between thinking and shipping
Instead of the usual chain of strategy → brief → copy → dev → review → rework, much of the friction was compressed.
2. Decisions happen inside the real system
Not in an abstract doc, but inside the repo, page structure, and actual content architecture.
3. Visible iteration
This did not stop at recommendations. The changes were actually written, committed, and deployed.
4. Critical sparring
The agent was not used as a cheerleader. It was used to ask better questions:
- Who is the real customer?
- Which pages damage positioning?
- What is legacy vs. strategically important?
- Which content should stay, be reframed, or be removed?
5. Better balance between speed and judgment
The tooling accelerated execution. The human kept strategic control.
What this means for clients
For clients, the relevant point is not which tools run in the background.
The relevant point is:
- faster iteration on shop, website and connected systems
- less handoff loss between analysis, decisions, and implementation
- clearer decisions inside the real system, not in abstract docs
- less delay caused by recommendations that never get shipped
- a workflow that moves from finding to commit in one loop
Technology stack
The stack in this workflow included:
- OpenClaw as the AI agent for analysis and file edits
- GitHub as source of truth for code
- Astro as the website framework
- Coolify as deployment target
- Leantime for project organization
- direct chat-based approval and iteration
Where this way of working is especially strong
- live project iteration
- content refactoring across larger existing sites
- positioning work with direct implementation
- repo-native tasks where analysis + edits + commits belong together
- ongoing improvements instead of monolithic relaunches
What it does not replace
It does not replace:
- human judgment
- strategic accountability
- real business decisions
- prioritization by the owner or team lead
But it noticeably shortens the distance between finding and implementation.
Summary
This is how we work in client projects too: finding, decision, change in the real system, review, deployment – in one continuous loop. AI tooling helps us move faster from analysis to implementation. Responsibility for direction and quality stays with the team.
If you want to explore what this could look like for your shop and the systems behind it: