Every migration I’ve led started with two things: a Proof of Concept, then a Proof of Value. They’re the cornerstone. And honestly, they’re just the fundamentals of project management, done on purpose.

1. Find the motive. When you’re making these changes, start by understanding the motive behind them. There’s a business outcome that needs to be achieved, or somebody’s promotion. Somebody wants this for some reason. They may not tell you, but that’s probably the reason. It’s your job to get curious, uncover it, and turn that ambiguity into something concrete.

2. Write a hypothesis, not just a discovery session. Most people call this step a discovery session. I’d add a hypothesis statement. Think about a time you had an idea of what you wanted to do and why. You still had to test it and validate it to see if it was what you actually wanted. Adoption works the same way, whether it’s a tool or a new framework. It comes from stating the benefits, and from understanding where you’re coming from.

3. Put it in a project charter. Some people call it a PRD (product requirements document), a one-pager, or a business case. They’re different, okay? A project charter has a few key things:

• The scope statement: what the project will deliver, how you’ll know it’s done right, and what it won’t include. (The charter holds the why and the when. The scope statement adds the detail.)

• The people: the economic buyer, the purchaser, the sponsor, and the key people you need to stay connected with.

• The communication plan: how you’ll update people, with what, and why, so everyone knows where the information came from.

4. Answer the five questions that matter.

• Who are you?

• Who is responsible for the work?

• Who’s going to sign the check?

• Why are they signing it?

• What’s the motive behind signing it?

Put all of that in the document.

5. Go find the pain points. From that document, start reaching out to people to learn what they need, what they want, and what they have problems with. Those are your pain points, and you document every one you can. There are a ton of methods: journey mapping, interviews, focus groups. If you don’t believe me, check out the PMP 8th edition. These are tools in your toolkit, and the expert part is knowing which tool to use at the right time to get you to the result faster. AI can’t do that yet. I’m sure it will, just not today.

6. Build a go/no-go checklist. In the military we had inspections. Every item was a go or a no-go, and there was a pass/fail threshold. With 10 items, you passed at 7 out of 10. At 6 out of 10, you failed automatically. Do the same thing for your adoption or migration evaluation. Now you have a clear path and a clear scope. It stops scope creep, and it’s documented. Everyone understands what you’ll do, when, why, and how you’ll measure success. That’s really all it comes down to, especially on large migrations.

Why this matters more now AI can help you build something super fast, but it’s not always right. It’s getting better, and the better it gets, the more our writing has to keep up. With enterprise tools, the better you write, the better context the AI has when it draws on past work. If you follow this process, anyone who comes after you can read it and understand it. If they hand it to an AI, they can trust what comes back, because what you wrote was thorough and concise: AI-assisted, with a human as co-author. That’s how you move the needle in your direction.

Grab the fill-in template: The Migration Playbook (PDF attached).Change Management

Gsd Migration Playbook Template

221KB ∙ PDF file

Download

Download

Hit reply: what’s the one document you wish had existed before your last migration?

Read more

Six years ago after I transitioned out of the Army, I sat down with our Chief of Staff. He was kind to me, which I remember better than most of what was actually said, and he told me to read The First 90 Days by Michael Watkins. I read it that year alongside my PMP training books and I remember being struck by how much the two overlapped. My dad would call that the clue bird landing.

I picked it back up this summer. I was coming off a rough stretch, felt a little lost, and wanted something to ground myself in. So I started re-reading it.

I got to the section on adapting to a company’s culture: how power gets distributed, whether meetings are where decisions happen or where they get rubber-stamped, whether the place is consensus-oriented or process-oriented, how people get recognized. My first reaction was that I’d never thought about any of this.

Wrong twice

That was wrong twice over. I thought about all of it in the Army. The military decision-making process (MDMP) is essentially a framework for reading organizational culture. And I’d already noticed the overlap six years ago, sitting there with a Watkins book and a PMP study guide open next to each other. I just didn’t hold onto it.

What carried over

Veterans get told to translate their experience for civilian jobs, and everybody hears that as job titles. Squad leader becomes team lead. That’s the easy part. The frameworks came with you too. And apparently you can know that and still need to learn it again.

What I had to unlearn

The one habit I had to break getting out of the Army and going into tech: nobody cares about your rank. Your position and title matter, just not in the same way.

In the military you respect the rank. Someone earned it, and there’s a level of respect you show for that. It’s like seeing the title CEO versus junior sales engineer. When you see the rank, you have a certain expectation of that person, no matter their background or what they do.

That chain of command was hard for me to find, break and understand in corporate America. There isn’t a strict one. There are different types of leaders, and different types of organizations where it matters. That’s why understanding the organizational culture, how people actually work inside a company, is so important to learn and to stay curious about. It sets the tone for how you’ll fit in.

If you’re a veteran in tech: what’s the framework you brought with you and forgot you had? Hit reply. I read every one.

I don’t have a productivity system. I have a pile of tools that each fix one way my brain drops things. Here’s the pile, what job it does, and where it failed me.

Capture: so nothing lives in my head

• Plaud: a voice recorder I talk into on walks and drives. Talking is how I think. The transcript lands in my notes automatically. Plaud Note Pro or Plaud Note

• Wispr Flow: dictation on my laptop. I talk instead of type, and it cleans up the mess.

Think: an external brain

• Obsidian: my notebook. Wins, lessons, and connecting dots. Not where I track work.

• Excalidraw+: my whiteboard. When I’m learning or rehearsing something, a drawing beats a wall of text.

• Claude: turns my rambling voice notes into a draft in my words. It drafts; I decide.

• Readwise: pulls my book highlights into my notes so reading turns into something.

Do: nudges, not willpower

• GitHub Projects: the one record of what I’m working on. Every post gets a date set backward from when it goes out, so I know what’s due today without deciding.

• Apple Reminders: one list, every task has a time. If it’s not there with a time, it doesn’t exist.

• Apple Calendar: Family life lives there. Helps show reminders and other activities.

• Everyday: habit tracker. I tap to log; it never auto-completes for me.

• iOS Shortcuts: soft evening nudges (“did you read?”) that ask before marking.

Dad

• How to Talk So Little Kids Will Listen: acknowledge the feeling first, then redirect. Changed bedtime more than any app. (Amazon link)

What didn’t stick

• I built a time tracker in two hours and broke it on day one. It only works if the stopwatch is running. Instead, I just use a physical stopwatch so that I can time myself. And when I forget, I take a note, send it to my AI and update my calendar so that at least it was on my calendar when I do a review later in the week.

The rule underneath all of it: if a tool needs me to remember to use it, it’s going to fail. Each one here works because it catches me, not because I’m disciplined.

As an Amazon Associate I earn from qualifying purchases. I only link what I use.

GSD stands for Get Stuff Done. This is where I break down the systems I actually use to run work, content and family life with an ADHD brain. I write every letter myself and use AI to help me sharpen my writing, the same way I'd use an editor.

How the week works

• Sunday: a new video goes up on YouTube.

• During the week: short Notes here on Substack with quick wins, behind-the-scenes and things I'm testing.

• Friday: the letter. It goes deeper on last Sunday's video, with the parts that didn't fit, what broke and what I'd do differently.

What I write about

• ADHD systems. External-brain setups that work when your focus doesn't. Start with My weekly video took 7 hours. Now it takes under 3.

• Practical AI. Using AI as a tool, not a crutch. Start with An AI Will Flatter You Out of a Job and Your Books and Your Conversations Belong in One Brain

• Adoption. Teaching people how to use tools, one coffee and visualization at a time.

• Modern-day parents. Figuring out parenting with my wife in real time. Start with Nobody could hear me

• Veteran to tech. What carried over from the military, and what I had to unlearn.

If you only read one thing, read The week I didn't post. It's the most honest look at how this whole thing runs.

Subscribe free to get the Friday letter. Hit reply anytime. I read every one.

Best,
Carlos M.

This week I uploaded a video to YouTube with no sound. I didn’t catch it. My friends and family did.

I had exported the same video several times and kept clicking mute on the audio without noticing. It looked finished, so I posted it. Then the messages came in: nobody could hear me.

So here’s the new rule. Nothing goes up until I play the export back, start to finish, with sound on. It’s a small habit, but this week it would have saved me the whole problem.

Same four questions as every week: what I posted, what I learned, what’s new, and what’s next.

What I posted

• Two longer videos. Don’t Let AI Do All the Work (DaVinci Resolve + Claude) and My weekly video took 7 hours. Now it takes under 3.

• Shorts on the channel, including the DJI Mic 3 audio fix.

• Last week’s letter, here on Substack.

• A new playlist idea that held. My wife and I talked about it about a month ago. This week it became a real playlist, Modern Day Parent, and we recorded the first episode. We found a few kinks, so we’re recording it again with fresh sleep, eyes, and coffee.

The goal I set is 90 minutes from an idea in my head to a finished video. I’m not there yet. Recording takes about an hour. Editing is still the slow part.

What I learned

Trust the check, not the feeling. The silent upload felt done. It wasn’t. Playing it back is the check, and it takes the length of the video. That’s cheaper than taking a video down.

My board was right and I was wrong. I keep my plan on a simple board with due dates. A few things I thought were finished were still open, and one was past due. It was a quick gut check: go back, fix it, update my notes. The board did its job, which was to hold me accountable.

Ideas after the edit cost the most. This week’s video took 18 versions to get right. A few were my mistakes. Most were new ideas that showed up after I watched it. Next week I want the graphics list written before the first cut, and three passes: the cut, the graphics, the final.

What’s new

• GSD Media is official. The state approved the name. It’s not just GSD Media Productions LLC on paper anymore. I can use GSD Media.

• No more website. Everything lives here on Substack, with its own web address. One place for posts, pages and notes is easier for my wife and me to keep up.

• Modern Day Parent. The more parents I talk to, the more I want to share what’s working for us. That’s the playlist we’re building together.

• A board I can actually see. My plan lives in five views: a roadmap, what’s past due, what’s happening now, what’s next, and what’s parked for later. Ideas I record as voice notes on my walks and runs with my son land on that list by themselves, so nothing in my head gets lost.

What’s next

• The first Modern Day Parent episode, with my wife, this week. You’ll get to meet her. She’s funny, a little quirky, and the reason this playlist exists.

• A “who am I” episode. I’ve been talking to you for three weeks without really saying who I am or what I bring to the table. That’s next.

• Three versions, not eighteen. Same video quality, fewer rounds.

The coach in the passenger seat

It’s been three weeks of editing in DaVinci Resolve with Claude helping me. Some creators let AI make the whole video. I’d rather learn the craft.

I used to own a six-speed turbocharged Mini Cooper. I loved that car. You had to work it, and that was the fun. Using Claude feels like having Lewis Hamilton in the passenger seat, giving me pointers while I still drive. Every week I learn something, apply it, and get a little better.

Not holding yourself back, and always getting better. That’s the message.

See you next week.

On Oct 5 I asked whether a recut could beat 3 h 52 min from footage to a published Short. I promised the time on Sunday and missed it. Running two ideas in one week put me behind.

The answer is 2 h 50 min, 62 minutes faster.

• What I learned: without a plan, I fall behind.

• What was new: Episode One, my first full episode.

• What’s next: one idea per week, and a goal of 90 minutes.

Full Review on the channel

Leave a comment

Why the change?

It was Wednesday, I was coming home from dropping off my son at daycare. I get home, my wife asked me “what are you working on today?” I looked at my Obsidian vault and I was like “I don’t know, I’m not sure what I’m doing.” Then I decided to pivot my system. I decided to look into why it wasn’t working for me. I had too many touch points in my digital system. It was Obsidian, Apple Reminders, Apple Calendar, Mail, GitHub and two more repositories I needed to manage. That was just too much.

Why I moved

I started this system in cloud-video-pipeline on September 1, and on September 25 I moved away from it because the work from the repo to my Obsidian vault was complicated and I didn’t like that flow. It was too difficult to find what I was working on next. I switched because it was too confusing to get the information out of my head and into a system I could see and follow.

The claim: the operating system is the bottleneck

Your audience isn’t the bottleneck, your operating system is.

I thought about this because most often in my experience running a weekly demo at my previous company, it was never the audience. They knew what they wanted, expected to hear, and what to expect. It was the presenter, the person eager to share who needed to follow a system. To help them show up as well polished professionals, that they are.

What I built instead

Twenty-three pull requests merged across three repos. A new logo, which my wife and I worked on together. A website that matches what we want more closely. A nine-stage workflow for how I capture content, scale it, test it, review it, pressure it and build on top of it.

What surprised me the most was how fast I could pivot and change everything in my entire system, from the website to the repository to the ideas.

I also don’t have to write ideas down on paper in the moment, which can feel rude in some settings. I say a quick word or two into my device, lean on my capture system, and trust the process I’ve built to help me build on top of it. It’s no longer scary to have a bunch of ideas and not know where to put them. I’ll still carry a notebook or an iPad to jot something down in a pinch.

The clock

I timed myself on an iPhone timer, idea to publish.

• Last week: 7 hours. A gear review. It was my first time learning Resolve, and I was using Claude.

• This week: 3 hours 52 minutes. A weekly wrap-up, with Resolve already learned and a better understanding of the tools and what they can do.

These are two different jobs, so I’m not claiming the second is the same work done faster. Part of the gap is that I’d already learned the tool. What I can say: editing is getting faster, and the ideas are becoming clearer.

• For me, it was about being able to measure myself and make sure that I’m holding myself to the same bar that I would hold anyone else to. If it can’t be measured by me and I’m not being strict on that, then I’m missing the point. That’s what I want to do: be honest in my claims and be truthful. It was proof to pressure test and stress test the system to make sure it was working as expected.

What I skipped, on the record

I skipped the week’s Short, Reel and TikTok, and the long. They are written on the board as skipped, on purpose. A missed deliverable stays as evidence, not fake success.

What I learned from the week prior to posting my first review was that learning a new tool is never easy, but with the help of AI, it’s something that you can do much faster. Understanding the basics, I think, of video editing and the fundamentals helped in that department. I also learned that just because a tool exists doesn’t mean you need to use it. It’s about helping you measure the outcome that you’re after. That’s more important. What I would do differently going into the next week is spend a little bit more time evaluating my skills and agents that I have set up in my current system. I would make sure that I have the right design for handing off tasks to different agents and building my human decision of what I want to focus on. I also learned that I want to get better at editing and faster editing. I also learned that I need to smile more on camera. I tend to be a little nervous for some reason, something that I was never like before. Now I think, because I’m doing this, there’s a little bit of nerves kicking in, but that’s something I’ll have to work on.

What I’m building: the muscle

What I want to post, and how I want to show up, is the new muscle I’m building. Writing, posting and testing my own system is the proof point.

Last week’s long and Short are here if you want the context: the Plaud long and the Plaud Short.

This week’s video: I skipped this week’s video on purpose. Here’s what I built instead.

What’s next

Starting October 5, I’m running a three-week test of the system, through October 25. I set two lines in advance. If a week runs over 110 minutes, I name the next rule change. If three weeks come in under 90 minutes with a long video posted, the narrow version of my hypothesis holds.

Next week I need to film, under a busy family week, not a quiet one. I’m refining the filming and editing process, and I’ll keep refining the research process in my repository.

Step my game up with getting more efficient at editing with Resolve. This week, I was able to edit the video using 40% of my Claude usage limit in one session of chatting. I also learned that I needed to get my mic timecode setup for better audio sync with my camera going forward.

Who this is for

This is for people who are struggling, who have ADHD, are super busy and can’t keep up with their system, and need a way to leverage AI effectively. If your system already works, you can ignore it.

Subscribe to the newsletter and follow the three-week test.

Seven hours. That’s what it took me to make a ten minute video this week, on a tool I’d never opened before Monday. For context, the professional benchmark is 45 to 60 minutes of editing per finished minute, which means a ten minute video is 7.5 to 10 hours of editing before you’ve filmed a single frame [1]. My seven hours covered everything: filming, A roll, B roll, photos, and the edit.

Here’s what this post covers: the problem (making videos solo takes too long and I was doing it the hard way), what I learned (a pipeline mindset plus a new editing tool got me under the industry editing benchmark in week one, with production included), and the way forward (why every week from here should get faster).

The problem

The math above is the wall every solo creator hits: the work is doable once, but not repeatable at a pace that builds anything.

My version of the problem had a second layer. I was over engineering. Too many connected tools, too many systems, too much ambition for one week. And there’s a tempting shortcut everywhere you look: AI can run a faceless content channel for you now, no face, no thoughts, no you. You can do it. I’m not going to. It just feels wrong to me. We’re people, and I want to talk to people. So the constraint I set for myself: AI helps me go faster, it doesn’t get to be me.

The learning

The fix wasn’t a tool. It was treating content like a pipeline. I was a product manager for CI/CD and I’m a PMP, so this is home turf: correlate work to stages, test it, put quality gates around it before you let it rip. It’s the writing process. It’s the scientific method. The prompt is your first draft; the revisions are still on you.

Here’s the pipeline I ran this week:

The tool that made the pipeline real was DaVinci Resolve [2], which now ships AI tooling that can rough in a first draft of your edit. I’m a Final Cut Pro guy, not a master but comfortable, and I had never opened Resolve before Monday. What made it click, fast:

• Keyboard shortcuts map almost exactly to Final Cut. Command Shift Bracket for ripple trims, Command B for blade. Your muscle memory transfers.

• Slip trims. T then S, then comma and period to slide content inside the clip. Small thing, huge time saver.

• Markers are your best friend. The aha moment. Color, title, notes, keywords on every marker. I built a tag system: one color for cuts, one for reusable clips, one for “doesn’t fit the story yet.” My timeline became readable at a glance.

• Auto Color across the whole timeline. Select all, apply, move on.

• Publish straight to YouTube from inside Resolve. No YouTube Studio step. Final Cut doesn’t have this. Win.

Two editing sessions, three and a half hours each, video done. But here’s the honest shape of the week: the setup cost was real, and I paid extra for over engineering it.

The way forward

Seven hours, week one, all-in, on a tool I’d never touched, against a 7.5 to 10 hour editing-only benchmark. That’s not the win. That’s the baseline. The whole point of building a pipeline instead of just grinding is that reps compound: the shortcuts are in my hands now, the marker system is built, the publish step is one click. Next week doesn’t start from zero, it starts from here.

Concrete fixes going into next week: shoot B roll deliberately instead of remembering at the end, keep the posting model running (a marketing video my wife shared frames every post as one of four types: attraction, nurture, conversion, or connection, so a Short early in the week to invoke a response, then two or three posts from my own ideas), and stop adding tools until the current ones are boring.

What I told you

The problem was speed and over engineering. The learning was that a pipeline with one human quality gate beats both grinding and going faceless: my full production came in under the editing-only benchmark in week one on a brand new tool. The way forward is reps, because a system you’ve paid the setup cost on only gets cheaper to run. Next week: faster.

References

[1] Veedyou, “How Long Does It Take to Edit a Video?” — https://www.veedyou.com/how-long-does-it-take-to-edit-a-video/
[2] Blackmagic Design, DaVinci Resolve — https://www.blackmagicdesign.com/products/davinciresolve

1. Conclusion

Two questions keep coming up when people find out I am running my job search with AI.

What is your setup? And how are you handling the bias?

The second question is the better one. An AI will happily tell you that you are a perfect fit for a role you are not qualified for, and then write you a résumé bullet claiming a 40 percent improvement in something you never measured. That bullet gets you a first-round interview and ends your candidacy the moment someone asks a follow-up question. The tool is biased toward making you feel good and toward sounding confident. Left alone, it will flatter you straight out of the process.

I am also biased. I am a platform product manager, which means when I am anxious my instinct is to build a systeminstead of making the phone call. Building feels like progress. It is not.

So my setup is designed around both of those biases rather than around the AI being clever. One strict written prompt that forbids unsupported claims and forces every detour to be named. One list of things to do. One schedule. Nothing new to maintain.

How is it working? Honestly, it is doing exactly one thing well: I no longer waste time deciding what to work on, and I no longer send anything out that I cannot defend in a live conversation. It has told me to skip roles. It has told me I was busy instead of productive. The interviews are still mine to win.

And the system is not the answer. It is the thing that clears the desk so I can do the actual work, which is talking to people. More on that at the bottom, because it is the highest-leverage hour of my day and it has nothing to do with software.

2. The Setup

The core is not a clever prompt. It is a strict one. A few pages long, doing four things.

• It defines the mission narrowly. Roles where my actual experience applies: product management, developer platforms, developer experience, internal tools, product intake and prioritization, technical customer engagement, AI-enabled workflow automation. Optimize for quality applications, warm conversations, proof of work, and interview prep. Explicitly not for application volume.

• It forces a scoreboard before anything else. Live processes, applications submitted this week, pieces shipped this week, deliverables due today, on-time status, blockers, recommended next action. I cannot open a session and chat. I get graded first.

• It has an evidence standard. This is the bias answer, and it is the part I would steal if I only stole one thing. No inventing, inflating, or implying any fact not supported by my materials. When a claim is plausible but unsupported, it must be labeled one of four ways: needs confirmation, evidence gap, potential positioning angle not yet substantiated, or do not claim. Forcing the model to flag what it does not know turns it from a confident liar into a useful editor.

• It has a focus protection rule. Any time I propose a new tool, dashboard, automation, or template, it must be named as a possible detour and challenged with one question: what shipped this week, and what active deliverable will this replace or accelerate? That rule exists specifically to catch me.

There is also a challenge rule. If I am on low-leverage work while something is due or overdue, it says so in four lines: observation, opportunity cost, recommended decision, smallest next action. No lecture, no shame.

3. Project

The project is a single Claude project I created on September 8, about a week after the layoff.

What lives in it:

• Résumé versions. A baseline résumé plus a tailored version per role, all in Obsidian under /projects/career-ops. Plain markdown files on my own disk. Not twelve files named Resume_Final_v3.

• Role dossiers. One per live opportunity: the job description, my evidence map against it, prep notes, and what was actually said in each round.

• Canonical figures. One file of the numbers and scope claims I am allowed to use, with a source for each. Everything else gets flagged.

• Interview debriefs. What the interviewer actually said, scored honestly, written within an hour of hanging up.

Hand it a job description and it does not write a cover letter. It runs a fixed workflow:

• A recommendation first: pursue now, pursue if a warm path exists, pursue only with targeted positioning, monitor, or skip.

• A fit assessment.

• An evidence map, as a table, mapping each role requirement to real evidence and labeling the strength direct, adjacent, or missing.

The missing rows are the point. It is not allowed to hide a gap behind vague wording. It has to tell me the gap is there and whether it is fixable through positioning, a portfolio piece, a referral, preparation, or not fixable for this application.

4. Schedule

The cadence exists so that showing up is not a daily decision.

• Weekday 7:00 AM Central. A morning brief runs on its own. Scoreboard, what changed overnight, what is due today.

• Friday 3:30 PM. A pre-retro brief runs.

• Friday 4:00 PM. A ten-minute retro. Wins, pipeline changes, applications, conversations, interviews, pieces shipped, what got missed, bottlenecks, lessons, and next week’s three priorities. That snapshot is stored so Monday does not start cold.

• Monday 9:00 AM. A fifteen-minute plan.

Underneath, the working day is about six protected hours, split roughly three for the job search and three for GSD Media, my creative business.

I want to be honest about the failure mode. The schedule is easy. The six protected hours are the hard part, and I do not hit them every day.

5. Context and Tool Use

Almost none of this is new software. It is what was already on the machine, connected so it can be read and written.

• Apple Reminders is the operating list. One list, called Deliverables. Every agreed task goes in with a specific due time. Checking the box is the only way completion is confirmed. The rule that keeps it honest: a deliverable does not exist until it has an outcome, an owner, a due date, a due time, a definition of done, and the source material needed to do it. If it is not in Reminders with a time, it does not exist.

• Calendar holds the timed blocks, so six hours is visible rather than aspirational.

• Mail and Calendar connectors let the morning brief see what came in overnight. Recruiter replies, scheduling requests, follow-ups I owe people. That is how what do I owe people today gets answered without scrolling an inbox.

• Obsidian is the source of truth for career evidence, in /projects/career-ops. Because the evidence lives in files I control, the evidence standard has something real to check against instead of the model guessing.

• Plaud, a small voice recorder, captures calls. The transcript becomes a written debrief while it is fresh, which is how what the interviewer actually said became a file instead of a fading memory.

• Excalidraw for thinking in boxes and arrows, usually mapping how a product’s loop works before an interview.

One hard rule across all of it: the system never sends anything. It drafts emails and creates reminders. Every send, submit, and apply needs my explicit approval on that specific item, not blanket approval. I am not comfortable with an agent emailing a recruiter on my behalf, and I do not think you should be either.

The part that is not about productivity

The three hours on GSD Media are not a side hustle line item. They are how I cope.

A layoff takes the thing you owned away from you. Interview processes then hand you weeks of waiting on other people’s decisions, which is the same feeling again in slow motion. GSD is the project nobody can take from me. I own it, I manage it, I learn on it, and I put my actual skills to the test on it. When a process goes quiet or a rejection lands, there is still something moving that week.

Everyone copes in their own way. Mine is keeping momentum on something I control. That is the reason a piece ships every week, including this one.

The next best thing you can do is network like you are speed dating, with better taste

The system exists so I have time for this. This is the part that actually moves a job search.

Speed dating gets the pace right and the depth wrong. Most people run one or the other badly: either a hundred cold connection requests that go nowhere, or three coffees over two months. You want the tempo of the first with the judgment of the second.

That means doing the homework before you ever send a message.

• Go through your connections. Then their connections. Then theirs. Slowly, on purpose.

• Build a mental model of who you know and who they know. Not a spreadsheet. A map in your head that you can actually recall in a conversation.

• Cluster them. By company, by function, by era of your career, by who tends to know everyone.

• Then go get coffee and be Sherlock Holmes about it. Who might know the hiring manager. Who sat next to them two jobs ago. Who ran the team that the team you want reports into.

Enjoy that part. The detective work is genuinely fun, and it is the difference between asking a stranger for a favor and having a real reason to talk to someone.

Quality beats quantity, every time. A handful of meaningful relationships does more than two hundred connection requests, and it does it in both directions. The person on the other side gets something out of it too, which is the only reason it holds. Do it this way and you spend far less time doing mental gymnastics to justify why a stranger should help you, because they are not a stranger and you are not asking for a favor.

One last thing

If you are in this too, the part I would actually pass on is this: take control of the time.

Not in the productivity sense. In the sense that the waiting is going to happen whether or not you do anything with it, and you get to decide what fills it. Build something. Learn something that is yours and not a line on a résumé. Keep one thing moving that nobody can take away from you.

And do not give up. People are still hiring people, despite the wave of AI. The conversations are still conversations. The person on the other side of the table still has to decide whether they want to work with you, and that has not been automated away.

Go do the homework. Go get the coffee.

If you have questions about any of this, the setup, the prompt, the evidence rules, how I am running the week, reach out. cvrlos. I will answer.

I spent the weekend building six finished cuts of that take, each on a different stack, and checking them. Every one rendered, passed its checks, and was a real video I could have posted. I put all six on Cap and sent them to eight people.

I put all six on Cap and sent them to eight people.

• v01, long cut (1:59, HyperFrames)

• v02a, Astra cut (1:55, HyperFrames)

• v03, mid cut (1:20, HyperFrames)

• v02, tight cut (1:00, bespoke build script)

• v04, Obsidian workflow (1:00, capture, Astra infographics)

• v03w, recordings to Obsidian (0:46, capture)

I picked v02. It is posted as a Short.

Why v02

v02 had the mix I was going for: graphic content with my own imaging. It looked like my work. That was the decision.

The cut some people liked

v04 was the most produced of the six. Astra infographics, a pass over the Obsidian workflow, three rounds of revision, an independent QA pass. On paper it was the best-made video in the set. Some of the eight people liked it.

I still did not post it, it just didn’t feel like me.

It was too fast, with awkward pauses in between, and the video did not make sense. The look was not the one I was after either. v02 mixed graphics with my photographs and the real screen. v04 did not.

Eight people did not pick the video for me. They told me a cut I would not ship still landed for some of them. I still had to decide whether that look was mine, and whether the film worked. It was not, and it did not.

What the weekend actually cost

Not the renders. Every stack in that table worked. HyperFrames, the bespoke build script, the capture path: all of them produced finished work. If execution had been the constraint, six cuts would have been six times the output.

The weekend went to watching them, sending them, and sitting with a cut that had fans and still was not the video. Putting the variants in a browser is now a method chapter. The part I would have skipped: stop revising a cut before you have decided it is the right cut. r02, r03 and r04 of v04 were careful work on a video that was never going to ship.

A note on the tool

The reason all six are watchable above is Cap. The pipeline publishes to it with one command, which is the only reason putting six variants in front of eight people was cheap enough to bother doing. Referral link.

My website

x.com

Let me give you the whole build, then the payoff.

Step 1 — Turn on AutoFlow. In Plaud Desktop, enable AutoFlow. Select your transcript meeting and choose a template that will transcribe the content and send you an email.

Step 2 — The email is the document. The body of that email is what was transcribed and summarized — the summary and the raw transcript as one document. That distinction matters. Summaries are lossy; the raw transcript is the source of truth. I’m using Gmail, but any inbox works.

Step 3 — Set the Mail rule. In Apple’s native Mail application, create a rule that looks for emails from that sender. Filtering on the sender makes this foolproof, rather than having it generically pull content.

Step 4 — Let your LLM write the script. Ask Plaud or your LLM of choice to write an AppleScript that automatically grabs the contents of the email — both the summary you configured and the raw transcript — and sends it to a designated path in your Obsidian vault. You don’t need to know how to code. You just need to know what you want.

From that point on: you’re driving home, you’re on a walk, you’re listening to a book and an idea hits. You record it on your Plaud, and that’s it — as long as your MacBook or Mac Mini is connected with Mail open, every summary and transcript flows into your vault without you touching anything.

Now the part that actually changes your thinking. Connect the Readwise MCP. As you read books and capture highlights, that content enters the same system as everything you’ve said and heard. Open Claude Code or Cowork in your vault, reference any files, and the LLM can draw a line between what you’ve read, what you’ve said, and what you’re thinking — connections you simply weren’t making before.

And it works in both directions. You can validate what was actually said in a transcription — not what you remember being said. You can test an idea from your commute against a passage you highlighted three months ago. I believe this is what makes an LLM a genuinely better thinking and sparring partner: it’s not reasoning in a vacuum, it’s reasoning with your sources.

We’ve got more projects, more teams, more deadlines. Not because you can’t keep track on your own — but if you’re a busy parent working as a professional, this is what you need: a simple flow that is set it and forget it. That way, if you ever have questions, you can go back and check, or test your own thoughts against what was actually said.

Honestly, if you haven’t done this already, I feel like people have fallen behind. But sometimes modern-day problems require modern-day solutions, without overcomplicating things.

We’re not there yet. But we’re close enough to feel what that kind of optionality actually means. For our family, it means room to make decisions without panic. It means more resilience if one of us gets laid off, needs a break, or wants to spend more time at home with our kids.

This is the story of how I stopped spending entire weekends on spreadsheets and started getting a clearer financial picture over Sunday morning coffee — using Claude Cowork.

The goal was never just money

Financial independence sounds abstract until you attach it to a real decision.

For us, the win was never a number on a screen. It was optionality: the ability to make family decisions with more freedom and less fear. That mindset — being financially informed, not just financially comfortable — is what made our monthly review feel non-negotiable, even when life got louder.

What the old process cost us

For years, our financial review followed the same rhythm:

• Friday: Download credit card and bank statements

• Saturday: Build the spreadsheet, categorize spending, sanity-check the numbers

• Sunday: Put together a short presentation for my wife — where we spent, how we were trending, what was working, and what wasn’t

It worked. We did fine. Before kids, losing six to eight hours over a weekend felt like an investment in ourselves.

Then we became parents.

You don’t know how much time you actually have until you have children. Someone told me that when I was eighteen. I finally understood it in my thirties.

Before kids: “Want to do this today?” “Nah, tomorrow.”

After kids: Has he eaten? Has he slept? Are we potty training? When did we last feed the baby? When did we change the diaper? If we travel, how long are we gone? What’s the pump situation? What needs to be cleaned before we leave?

Parenting turns your household into a mini logistics operation. If you’ve ever gotten a toddler out the door, through an event, and back into bed on time, you already know how to run a project with moving parts, tight deadlines, and zero tolerance for dropped balls.

The financial review didn’t become less important. It became harder to protect the time it required.

Why most finance apps weren’t enough

I tried the obvious route first: connect your accounts, let the app categorize everything, glance at a dashboard.

That part is fine. Aggregation is solved.

What’s not solved is the story.

Most financial apps tell you what happened. They rarely help you understand what you’re doing well, where risks are building, what decisions need attention, or how today’s spending connects to a goal six months out. Even good apps track and categorize well, but they often stop short of helping a household run a real review together.

I’m a product manager. I like timelines. I like visualizations. I like stakeholders being informed — and in this case, my wife is the most important stakeholder in the room. I wanted something closer to a project status report than a balance sheet.

That’s the gap I kept hitting.

I started treating our finances like a project review

Once I reframed the monthly review as project management instead of bookkeeping, the structure became obvious.

A good status report gives stakeholders a recurring snapshot of health against a baseline: accomplishments, schedule, budget, risks, issues, and clear asks (PMI). Our household finances follow the same shape.

Project management concept Our financial review Baseline plan Savings goals, vesting schedules, fixed obligations Schedule / milestones RSU vest dates, tax deadlines, big purchases Budget actuals vs. plan Monthly spend by category Risks Upcoming cash crunches, lifestyle creep, concentration Issues Categories already over budget Decisions / asks “Do we pause this subscription?” “Can we afford this trip?” Stakeholder My wife — the person who needs clarity, not raw data

The output I wanted wasn’t a chart. It was a decision-ready artifact we could read over coffee in five minutes.

Building the review in Cowork

When Claude Cowork launched, I started experimenting with whether I could compress the review without losing the quality of the conversation.

The first attempt was scrappy. Some formulas were wrong. Some categories were off. But the whole first pass took about forty-five minutes, and that alone told me the direction was right.

Over two weekends of focused build time, I got serious.

I created a Cowork project — a persistent local workspace with its own instructions, files, memory, and scheduled tasks.

On the data side, I built a simple pipeline around the same ETL pattern finance teams use in corporate environments — extract, transform, load.

• Extract — Download statements and CSV exports from our institutions

• Transform — Clean and normalize the data so categories stay consistent month to month

• Load — Feed everything into the project so it can be reviewed against our baseline plan

One concrete example: different institutions export similar data with slightly different labels, date formats, and category fields. Before review, I normalize those columns so the same type of transaction lands in the same bucket every month instead of creating false changes.

The inputs aren’t exotic. They’re just complete:

• RSU statements

• Pay stubs

• Checking account statements

• Credit card statements

• Brokerage and retirement account statements

Under the hood: the rules that keep the AI honest

The pipeline is the easy part. What makes the system trustworthy is a small set of data-integrity rules I wrote down once and now enforce on every session:

1. One source of truth. A single profile document holds every financial fact — income, obligations, accounts, goals. Every session loads it first, and it’s the only file that mutates. Everything else is a dated, immutable snapshot.

2. Every number carries a verification flag.

   • ✅ verified against a document

   • 🟡 estimate

   • ❓ unconfirmed

   The AI is not allowed to silently upgrade an estimate to a fact — a flag only changes when a source document says so.

3. Recompute independently. The AI re-derives the math instead of trusting a number because it’s written down. Twice this has caught errors I would have carried forward for months.

4. Hard guardrails. The AI never moves money, never executes anything, and never recommends a financial action without showing the trade-offs — pros, cons, and tax implications, with the numbers.

5. A reusable prompt library. Monthly review, cashflow forecast, tax estimate, “what if” scenario modeling — each one a documented, repeatable analysis instead of a fresh conversation I have to steer from scratch.

If you’ve ever written a playbook or a team working agreement, this is the same instinct. The AI is a capable new teammate; the rules are its onboarding doc.

What comes out the other side is a single HTML artifact — three tabs, clickable, clean enough that my wife and I actually want to open it.

Here is a sample of the flow

It answers the questions we care about:

• Where did the money go?

• Which categories are creeping up?

• Where are we strong?

• Where should we set alerts, the way you’d set monitoring thresholds in software?

• What opportunities are we missing?

• What weaknesses or threats need attention this month?

Not generic budgeting advice. Our story, told with our data.

The new ritual

Here’s what the workflow looks like now:

• Friday night: Download the month’s files and drop them in a folder on my Mac

• Saturday: A scheduled Cowork task processes everything; I do a quick QA pass to make sure nothing was missed

• Sunday morning: Before the kids are up, over coffee, we review one artifact together

Total review time: about ten minutes. The downloads and the Saturday QA pass add a few more, spread across the weekend in gaps I’d lose anyway.

Compare that to the old six-to-eight-hour weekend.

The time savings are real. But the bigger shift is psychological. We’re not asking “Are we okay?” from a place of anxiety anymore. We can see the trend. We can name the trade-offs. We can celebrate what’s working and flag what needs attention without turning money into a fight.

What actually changed

Three things stand out.

1. The review tells a story, not just a total. We see strengths, gaps, opportunities, and threats in the same view. That makes the conversation productive instead of defensive.

2. It’s built for us, and it can grow with us. New account? Add it to the folder structure. New goal? Update the project instructions. We’re not waiting on a product roadmap.

3. It respects the season we’re in. I’m not looking for a fancier finance app. I’m looking for clarity without sacrificing the weekend. With young kids and a household that runs on routines, that matters more than any feature list.

The takeaway

If you’re already doing the work of being financially informed — downloading statements, tracking spending, trying to align daily decisions with long-term goals — you may not need another dashboard.

You may need a system that helps you review your finances the way you’d review a project: risks, trends, decisions, timelines, and a narrative your partner can actually engage with.

I invested two weekends upfront to build that system. Now I get that time back every month, with a better conversation at the end.

If this sounds familiar, here’s the before-and-after at a glance — and if you want to build your own, the structure is simple: one folder, one instructions file, one scheduled task, and a Sunday morning to read the result.

How do you run your household financial review? Reply and tell me — I’m especially curious if anyone else has hacked together their own system instead of using a budgeting app.

Carlos Martinez is a product manager and builder based in the U.S. He writes about using AI tools for personal productivity, family logistics, and the boring systems that make ambitious lives actually work.

Further reading (optional links for curious readers):

• Claude Cowork product guide

• Project status report basics

• Preparing project status reports (PMI)

• The unbundling of the budget app

I bought the A7RV because I wanted a camera that would grow with my family for the next decade. I don’t care what people say, I am keeping this for as long as I can to get my ROI. But the megapixels per mile are definitely there. The only challenge (understanding this is a me thing), is that the camera is bulky. Period. It is AWESOME for shoots and anything else. It’s just bulky. I have tried running around with a smaller prime but it’s not the same.

So I was looking at the A7CR II and I liked it but there was something missing. I wanted something different. Like a completely different experience.

This could compare to saying you have a latte everyday, but one day you want to get “dirty”, and try the “dirty matcha latte” that your wife is raving about and then you get hooked. Well the Leica Q3 was my dirty matcha.

I bought this because we are going on a ten year awaited trip to Italy and I wanted this to be special. I also didn’t want to have to carry a mini studio with me while pushing a stroller, baby bag, and anything else. I just wanted that “it factor.” I found it in the Leica.

The Hold

The weight is lighter than my Sony, but damn it feels good to hold. Add in a little $40 thumb grip and we are cooking.

The Image

The photos are just sick man, the Leica color correction is something I can’t explain.

The Details That No One Mentions

After updating, I love that I can keep my profiles on my SD card — no one mentions that. Which is awesome because I am going to try and Claude Code some magic after my Italy trip to see what’s possible.

The menu button and play button are honestly a godsend. I love the simplicity in the menu. My favorite feature is that I can save 6 presets for my Italy trip that match my style.

The Wait Was Worth It

I waited 4.5 years to buy a camera like this and have always said “one day.” This last Tuesday was that “one day” and I am so pumped.