At a glance
Project intake ran through a form, a hand-typed ClickUp task, a hand-written project document, and a manual developer handoff.
- Automated ClickUp task and Google Doc creation directly from the intake form submission
- Reviewed each new project and recorded a video brief before assigning it
- Assigned developers from Telegram, which creates the ClickUp assignment and notifies them automatically
Roughly 30 minutes saved per project intake by estimate, not a tracked measurement, with dependency-related delays down by roughly 20% and the intake-to-assignment workflow running about twice as fast.
The situation
Marketing filled out a form for every new project, and from there the process ran by hand: someone typed the ClickUp task, someone else wrote the project document from scratch, and a developer got assigned in a separate conversation. [NEEDS INPUT: what else was competing for the same time that made this worth building] I built the automation to remove my own repetitive part of that chain.
How it surfaced
This was not a single bad week, it was a chain of small manual steps that repeated on every new project regardless of how busy the week already was. The workload that made it worth fixing was the same one already stretching the time available to fix it.
What I ruled out, and why
I looked at where the actual repetition sat: the ClickUp task was pure transcription, the same fields moved from the form into a task by hand. The project document was not: writing it meant reading what the client actually needed, doing whatever research the brief required, and organizing that into instructions a developer could act on. That is requirement interpretation, not template filling, so that is the step I put an LLM on: it takes the form input, interprets the client's requirements, and drafts the project documentation, including the instructions section aimed at the developer.
The decision and what it cost
I chose to let the LLM handle requirement interpretation, because that step was already text in, text out, and repeatable, but I kept the step after it mine: what the LLM drafts does not go straight to a developer. I review it myself, record a video brief from it, and only then assign the work. That review step is not a gap in the automation, it is the design: a model can draft a document, but the call on what a project actually needs and who should build it stays with me.
What I did
A marketing form submission now creates the ClickUp task automatically. An LLM reads the same form input, interprets what the client is asking for, does the research the brief needs, and drafts a Google Doc from a documentation template I built, including the instructions section a developer would act on. That draft lands in my inbox, not a developer's. I review the project and record a short video brief from it. Assigning a developer happens from Telegram, which creates the ClickUp assignment on that developer and notifies them: no step re-types information the form already collected.
The outcome
Roughly 30 minutes saved per project intake, by estimate rather than a tracked measurement. Dependency-related delays are roughly 20% lower and the intake-to-assignment workflow runs roughly twice as fast, both estimates rather than instrumented numbers. [NEEDS INPUT: whether this was built during the Full Stack Developer role in mid-2025 or after moving into project management. The roughly one-year timeframe points at mid-2025, before the PM roles started, and should be reconciled against the /experience timeline]
What stayed changed
Not every repeated step is a transcription problem. Moving a field from a form into a ClickUp task is; drafting a project document from what a client actually needs is interpretation, and that is where I put the model, not on the parts that were already mechanical. I left every decision about what a project needs and who builds it exactly where it was: with me, reviewing what the model drafts before anything gets assigned.
Related incidents
Scaling Engineering Teams: Taking the Exprovia Engineering Team from 7 to 20
I built the hiring pipelines, onboarding flows, and mentoring structure that took the Exprovia engineering team from 7 to 20, spending senior engineers' shipping time on context transfer to buy capacity that compounds.
- Hiring Pipeline
- Onboarding
- Mentoring Structure
Sprint Velocity Dropped 40% Mid-Sprint: What To Do About It
A 4-person Exprovia team's output dropped 40% mid-sprint, with 15 days left against roughly 25 days of remaining work. The cause was not effort: junior developers were taking complicated paths through problems on the codebase's critical parts, where simpler ones existed. I sat with each of them individually, worked through simpler approaches together, and spent the intervening weekend getting ahead of it. Velocity recovered in about four days.
- 1:1 Coaching
- Rate-Limit Bug
- Sprint Recovery