Two tools competing.
The website used Next.js to organise its pages, GSAP to animate them and Lenis for smooth scrolling. The conflict was in how the animation tools worked together.
I build the process
into the tool.
I use AI to turn the way I think about construction, estimating and operations into tools I can use again.
My methods, saved as instructions. Calculations I can check. Tools built around the work.
See the systemsFrom expertise to execution
InputDates, documents, business rules
My skillThe method, made repeatable
ExecutionTools, scripts and connected data
ReviewAn output I can inspect
Human judgement stays in the loop.
Construction knowledge × Systems thinkingExplore the work ↓
Selected systemsThe work behind
The work behind
the workflow.
I write my own skills: reusable instructions and tools that encode a process I understand. These examples show how I apply that approach to construction planning and estimating.
Construction planningCustom skill / construction-programme
From contract dates
to a calculated programme.
A programme needs a defensible sequence, a working calendar and dates that reconcile. I built construction-programme to bring those requirements into one repeatable workflow.
It uses submission, award, site possession and Practical Completion as anchors, applies a library of construction logic, then calculates the dates by script.
Define the anchors
Start with the commitments.
Contract dates establish the boundaries. Calendar rules and assumptions make the basis of the programme visible.
Apply the logic
Make the dependencies explicit.
The skill brings construction sequencing into the workflow. A script calculates dates, so dependencies can be checked and changes traced.
Review the output
Produce something usable.
The output is an Excel workbook in an MS Project-style layout, including a weekly Gantt, critical path, calendar and assumptions, ready for review and submission or import.
Programme logicInteractive illustration
Critical pathAvailable float
Calculated completionWorking days
Simplified example, calculated in your browser. Monday to Friday, no public holidays. This demonstrates the logic; it does not run the full skill.
My contribution
Construction logic, skill design, scripted calculations and the structure of the final deliverable.
Estimating practiceReusable tools for everyday estimating
Make the method
available every time.
I’m building the estimating process at Beek Build from the ground up. Alongside the day-to-day work, I’m turning parts of that process into reusable instructions and tools that make analysis more consistent and repeatable.
Each skill has a defined job. Together, they support the way I review scope, research requirements and examine project risk.
sow-evaluator
Give scope review a repeatable structure.
A custom skill for evaluating scope of works as part of my estimating process.
InformationScope of works
MethodStructured evaluation
PurposeConsistent review
The judgement is still mine. The skill gives the review a method I can reuse.
My contribution
The estimating process itself, translated into specialised tools. This is an evolving working practice, with professional review retained.
A lesson from the buildDiagnose first.
Diagnose first.
Then change the system.
While building the podcast website, the scrolling and animations stopped working properly. Two tools were trying to control the same movement: GSAP, which animates the page, and Lenis, which makes scrolling feel smooth.
Rather than keep asking the AI to “fix it”, I changed how I gave it the task. I brought in specialist instructions, described the problem and required a diagnosis and proposed plan before any changes.
Bring in the right guidance.
I added specialist instruction packs for animation, scrolling and performance. These gave Claude Code guidance specific to the problem instead of another open-ended request.
Explain before editing.
I wrote a structured brief: investigate the cause, explain the proposed fix and wait for my approval before changing the code. That made the next step something I could assess.
Direct the work deliberately.
Using AI to build something still requires judgement. My role is to define the problem, choose useful guidance and understand the proposed change before letting the tool act.
Tools in everyday useWhat I use.
What I use.
What I use it for.
I started with ChatGPT, before I began using Claude, creating personalised assistants through its custom GPTs feature. I gave each assistant instructions for a particular task, so I could reuse the same approach.
That has grown into writing my own instructions, building software with AI and connecting tools to the information I work with. These are a few examples of my practice, not a complete list.
ChatGPT · ClaudeWork through a problem
Work through a problem
and make it clearer.
These are AI assistants I can talk to in everyday language. I use them to develop ideas, organise requirements and turn rough notes into a clearer brief.
ChatGPT was my starting point. Claude is now also a regular part of how I work, particularly when developing instructions for tasks I need to repeat.
In practice
Explaining how my estimating process works, then breaking it into clear steps that an assistant can follow.