Running an AI Music Project: Week 3 Progress Report
Clafacio Lobo · Project Manager, Musinique · Humanitarians.ai
Three weeks in. This is the week the project stopped feeling like a setup and started feeling like a system — one that is producing real output, carrying real risk, and requiring real decisions.
This report covers what we built, what broke, what I got wrong, and what I am carrying into Week 4 with specific accountability for each item. If you have been following along since Week 1, this is the week the honest version of the story starts.
What Changed Between Week 2 and Week 3
Week 2 was high energy and wide surface area. A lot of output, a lot of momentum, some coordination gaps that I flagged and said I would fix. Week 3 was the week I found out which of those gaps I actually fixed and which ones I just described fixing.
The answer is mixed. The publishing pipeline is stronger. The research workstream has a clear shape. The Ghost Artist content series is now established with both written and video assets across two contributors. But the operational layer — upload confirmations, escalation ownership, feedback that goes beyond direction — is still the place where the project leaks.
I am going to be specific about that, because a progress report that rounds up is not a progress report.
The Team’s Output This Week
The content output across the team this week was the strongest it has been.
Shruti has now published three Substack articles and completed the research paper pipeline scoping — three distinct papers built on the Musinique dataset, with target journals and timelines identified for each. The collaboration pitch to SubmitHub and Artist.tools is complete, revised, and ready to send. It is still held pending data from Artist.tools. That dependency is the biggest single risk to the research timeline right now, and it is outside our control.
Ragamalika completed the Ghost Artist content series for this phase — two articles, two YouTube tutorials, and an original track. She has also begun research into Spotify for Artists as the foundation for the next article. Early findings suggest the platform is more complex than anticipated and the article angle has not yet been determined. That is appropriate given the other deliverables she shipped this week.
Nidhi delivered the highest output volume on the team — three articles, two YouTube videos, and a produced Ghost Artist track. She also did preliminary research and scoping for the Ghost Artist social media account setup workstream. That workstream is fully blocked pending platform access that has not yet been granted. The scoping work was the right use of available time while the blocker is pending.
Nixon continued building out the Artist.tools documentation review and contributed to the collaboration pitch strategy. The Substack article is still in outline phase. The thesis is too wide and needs to narrow before the writing can move forward. I gave feedback on this last week. I am not confident that feedback was sufficient. More on that below.
Sakshi made real diagnostic progress on the Claude embedding issue at irreduciblyhuman.xyz. The leading suspects — CSP conflicts and iframe restrictions — are better understood now than they were at the start of the week. The fix requires a site architecture decision from Professor Nik. Until that sign-off arrives, Sakshi is working at the edge of what she can move independently.
What I Am Thinking About As PM
This is the section I want to spend the most time on, because it is the one that is hardest to write and most useful to read.
The gap between direction and enablement.
I told Nixon last week to narrow the thesis. That is correct. But I delivered it as a note in a check-in, not as a working session where we sit together and find the argument. Those are different things. Giving direction feels like helping. Sometimes it is just pressure with better vocabulary. Whether a complete draft arrives this week will tell me whether I actually helped or just told him what to do.
I have been thinking about this more broadly. A PM’s job is not just to identify what needs to happen — it is to make it possible for the right person to do it. Feedback without time, escalation without ownership, direction without presence — those are half-measures. They feel like coordination and sometimes they are not.
The upload confirmation gap — still.
I flagged this in last week’s report. I said I would build a standardized checklist and distribute it at the start of the week. What actually happened is that the checklist exists but the enforcement was not consistent. By end of week, multiple contributors still had pending upload confirmations. This is the second week I am naming this gap. Naming it once is honest. Naming it twice without closing it is a pattern, and patterns are what I am accountable for.
The Nidhi escalation — still no confirmed owner.
Nidhi’s Ghost Artist social media workstream has been blocked for the better part of two weeks now. I escalated the access issue last week. I still do not have a confirmed owner for the resolution. Escalating without an owner is not a resolution — I said that last week too. The difference this week is that I am naming a hard deadline and a stated consequence. If access is not confirmed by Wednesday of next week, the social media setup workstream gets formally removed from the Week 4 scope and the timeline impact gets documented. That is what accountability looks like when escalation without ownership keeps producing the same result.
Cognitive load is accumulating.
Three weeks in, the volume of context I am holding simultaneously is significant. Five contributors, four workstreams, multiple external dependencies, a research paper pipeline, a collaboration pitch strategy, technical infrastructure that depends on external parties. By Friday of this week I could feel the edges starting to blur.
The coordination dashboard is supposed to be the external memory that offloads this. It works when it is current. It was not fully current until Friday afternoon this week, which means I ran most of the week on internal context rather than a reliable external record. That is fragile. It is also fixable, and it is the first thing I am fixing at the start of Week 4.
The Research Collaboration Pitch — A Status Update
This is worth its own section because it is the highest-leverage external action the project is waiting on.
Shruti’s pitch to SubmitHub and Artist.tools proposes a data-sharing agreement and a potential co-authored research paper. It is well-framed, strategically sound, and ready to send. The SubmitHub pitch does not depend on Artist.tools data and will go out with confirmed sign-off early next week. The Artist.tools pitch has a data dependency — we are waiting on platform data to complete the results section of the primary research paper.
The trigger condition we agreed on: if the data does not arrive by a specific date next week, we send a modified version of the pitch that does not depend on it. We do not keep waiting indefinitely. That is the decision.
What Done Looks Like in Week 4
I am not listing next steps as intentions. I am listing them as commitments with done conditions.
Nidhi’s access gets resolved or the scope changes. By Wednesday I will have a named owner and a resolution deadline. If the deadline passes without resolution, the Ghost Artist social media workstream is removed from the active scope and documented as a timeline risk. Done is a binary outcome either way — not still pending.
The upload checklist is enforced, not just distributed. I will track confirmations directly rather than waiting for them to come in. Done is every contributor’s Week 4 uploads confirmed by Thursday.
Nixon delivers a reviewable draft. I am scheduling a working session — not sending a note. We find the argument together. Done is a complete draft in my inbox by Thursday.
Both collaboration pitches have confirmed send dates. SubmitHub goes out on sign-off early in the week. Artist.tools pitch gets a hard trigger condition. Done is confirmed send dates for both, not pending status.
My Substack article is published. The draft exists. It is not published. Publication deadline is Wednesday. This does not get deprioritized.
The coordination dashboard is confirmed live on the project site. I follow up directly Monday. If Dev the Dev cannot confirm by Tuesday I find an alternative upload path. Done is a live URL, not a pending confirmation.
What This Project Is Actually Building
I want to close with this because it is easy to lose in the operational detail.
Musinique is not just a project that uses AI to make music. It is an attempt to build a documented, replicable framework for how independent artists can use AI tools to navigate platforms that were not designed with them in mind. Every article, every Ghost Artist track, every research paper, every collaboration pitch is a piece of that framework.
The coordination work I do exists to protect that mission — to make sure the pieces connect, the gaps get closed, and the people doing the creative and research work have what they need to do it well.
Three weeks in, the framework is taking shape. The publishing pipeline is active. The research pipeline has a clear scope. The Ghost Artist content series is establishing itself. The technical infrastructure is moving toward resolution.
There is real work left. There are real gaps I have named here directly. But the project is alive and it is building something worth building.
More next week.
Clafacio Lobo is the Project Manager for Musinique, an AI music research project at Humanitarians.ai. Follow the project at musinique.net · humanitarians.ai/clafacio-lobo