This content originally appeared on HackerNoon and was authored by Vimal Dhupar
I don't write code for a living. But last month I shipped a production dashboard that my engineering team uses in their day to day.
I'm a Technical Program Manager. I run cross-functional AI infrastructure programs — capacity planning, resource allocation, model launch coordination.
Last quarter, I needed a capacity dashboard, but with data scattered across multiple systems, each owned by a different team, there was no unified view. I realized that the Leadership was making allocation decisions on stale and manual spreadsheets.
I filed a request with engineering, however it landed in a backlog of requests with an estimated delivery in next quarter. So I opened an AI coding assistant and built it myself.
What Vibe Coding Actually Looks Like
It's not asking ChatGPT or Claude or Gemini to write a script. It's a conversation where I bring the domain knowledge and the AI brings the syntax.
I describe what I need: "Pull GPU allocation data from this query engine, join it with reservation data from this other system, compute utilization by team, render it in a dashboard with filters." The AI generates code. I review it for logic, not syntax. Does this join make sense? Is this calculation right? When the output doesn't match reality, I know which assumption is wrong. Then I iterate. "This returns duplicates — deduplicate by latest snapshot per day." Or: "This runs in four minutes. Push the date filter into the subquery."
That four-minute query? Got it to five seconds. Not because I know query optimization theory. Because I know the data well enough to spot the waste, and the AI knows how to fix it.
Why This Works for TPMs
We have the domain knowledge AI lacks. An engineer writes clean code but may not know which capacity metric leadership cares about. The TPM knows what the program needs, which systems hold the data, and what the output should look like. That context is the hard part. The code is the easy part.
TPMs see gaps nobody else owns.
Every program has them — reports that don't exist, dashboards half-built, data only accessible through manual queries. Engineers won't build them — not on the sprint. The TPM sees the gap every day but could only file a ticket and wait. Vibe coding closes that gap. Not in a quarter but in a matter of days.
We don't need production-grade code. I build internal tooling — dashboards, reports, data pipelines for weekly reviews. The bar is "useful and accurate," not "architecturally elegant." Vibe coding clears that bar easily.
Where It Breaks
When you don't know the data model, you're guessing. The AI will confidently query a table that doesn't exist or join on a misnamed column. I've gone back and forth with the AI for hours and hours, both of us wrong, because I assumed a column name and the AI ran with it.
To fix this: stop guessing, look at the actual schema, come back with real names. Domain knowledge isn't optional. It's the entire foundation of Vibe Coding.
When the codebase is unfamiliar, you break things you can't see. I once made a change that looked correct in isolation but broke a downstream process I didn't know about. The AI couldn't warn me — no context on the full system. A senior engineer caught it in review. Lesson: vibe coding works best when you're building new, not modifying something complex you don't fully understand.
When the framework has opinions, the AI hallucinates. Internal frameworks have conventions that aren't in public docs. Wrong function signatures, deprecated patterns, type mismatches. I maintain a short pre-submit checklist. Skipping it guarantees a failed review.
When you need someone else's system, you need a human. The AI can write the integration code. It can't send the message asking for permissions.
The One Remaining Blocker
Every blocker above has a workaround. Learn the schema. Build new. Maintain a checklist. Build relationships.
But one doesn't have a clean fix yet: code review.
I build a working dashboard in a day. Getting it reviewed takes two weeks. The reviewing engineers are busy with their own sprints, and are also busy reviewing the huge influx of AI generated code.
The irony: AI made building ten times faster. The human process hasn't caught up. The vibe-coding TPM is bottlenecked not by her ability to create, but by the organization's ability to absorb what she creates.
This will change. New review models for AI-assisted contributions. Automated checks handling the mechanical stuff. The norm of "only engineers ship code" continuing to erode.
But right now, if you're a TPM who vibe codes — your biggest challenge isn't the code. It's the last mile.
Why It Matters
For fifteen years, I've known what the program needs, where the data lives, and what the answer should look like. For fifteen years, I waited for someone else to build it.
That wait is over.
The TPM who vibe codes doesn't replace the engineer. She unblocks herself. Closes the gap between "I know what we need" and "here it is" — in hours instead of quarters.
Not every TPM needs to do this. But every TPM should know that they can.
This content originally appeared on HackerNoon and was authored by Vimal Dhupar
Vimal Dhupar | Sciencx (2026-05-13T03:32:35+00:00) I’m a TPM. I Vibe Code, and it Changed Everything.. Retrieved from https://www.scien.cx/2026/05/13/im-a-tpm-i-vibe-code-and-it-changed-everything/
Please log in to upload a file.
There are no updates yet.
Click the Upload button above to add an update.