Year's End: Looking Back at 2025

2025 was a year of transitions, but not always change. It felt like a focusing, and in ways an acceleration.

The Career Move

Early in the year I decided it was time to move onto something smaller again, a fund or a startup, and I began slowly keeping an eye out. After an on brand mix of serendipity (saw a linkedin joke) and followup, I decided to accept a new role in August. In October I joined Scale Venture Partners as Chief Data Scientist. After years building ML systems for venture data at Aumni and JPMorgan Chase, I’m now working directly with a VC firm to embed AI infrastructure into their investment process.

The role fits a pattern I didn’t see until I looked back: building ML systems that augment expert judgment rather than replace it. From predictive maintenance at Predikto to venture data at Aumni, the through-line has been systems that surface the right information at the right time while remaining explainable and interrogable.

At Scale we’re building infrastructure for tracking exceptional talent, anticipating round timing, and detecting early market patterns. The technical challenge is working with limited historical data and long feedback loops. The cultural challenge is building tools that investors can trust and pressure test. Both matter.

In December I was suprised and saddened to hear JPMC was shuttering Aumni in the new year. Aumni did a lot to bring true rigor into the complicated and messy world of VC portfolio management and I’m proud of the influence we had on the industry. There are loads of brilliant people still there as of this writing that are out there looking for their next thing, so if you want an intro, let me know.

Rennovation

At roughly the exact same time as starting at Scale (because I am not very smart), we also began a pretty significant house rennovation. Our little corner of Atlanta has continued to grow and develop into a slice of heaven. The silver comet is connected to the whetstone creek trail and soon the beltline. The westside park is massive and adding a bike park. Standing peachtree park is finally getting some of the love it so desperately needed from the Chattahoochee riverlands project, and the city is purchasing the Galloway fields for even more parks in the area.

I quite like parks and bikes, and the more an area has the more I want to stay there.

Home Life

At home things have been wonderful. I’ve enjoyed being a dad more than ever and as my daughter gets more capable and active we have more fun. She loves going on bike rides, “adventures”, playgrounds, and just chilling. She’s a blast. Our little family traveled to 4 countries and 10 states this year all together, including a wonderful thanksgiving in Barcelona with friends.

The Open Source Work

Kept maintaining [pygeohash], [elote], [keeks], and others. Nothing dramatic, but consistent updates and bug fixes. The libraries keep finding users, which is gratifying. Both [category_encoders] and [pygeohash] are apparently top 1% of python packages all time, with [pygeohash] having recently crossed over 100M all time downloads. Crazy.

I also published several MCP servers: todolist-mcp for task management (later became ContextSwitch), makefile-mcp for running build commands, and various others that made their way into my daily workflow. Some CLIs like [stargazers] and other odds and ends.

The Personal Software Year

As I gained more and more experience with AI-assisted coding this year I was able to get back into the habit of building new tools for myself (evergreen, contextswitch, lexicon, meadowview, betaday, idealog, speccheck, etc). These started as just things I used, but by year’s end I decided to actually distribute some of them. Evergreen is now on the Mac App Store for anyone who wants a simple, local CRM with MCP integration.

I’ve really enjoyed the way that these models have let me automate the boring stuff as it were, and scratch that itch. Putting something on the App Store felt like a nice capstone to the year.

The Writing Year

This site itself was revived in January, though I’ve been writing online in some capacity for close to 20 years now. I published 216 posts this year covering AI, decision-making frameworks, leadership, venture capital, the MCP ecosystem, mountain biking, parks, recipies, python development, and more.

My goal was really consistency, the discipline and ritual of writing, publishing, and seeking readership. And in that I’ve considered this a success. There are now many hundreds, even thousands of you dear readers most months, and growing. So cheers! I hope you enjoy this too.

Major themes:

Decision-Making Frameworks: Deep dives into Simon’s decision theory, recognition-primed decisions, nursing decision cycles, and the difference between data-driven and data-justified decisions. The common thread: how experts actually make decisions under uncertainty versus how we think they should.

Leadership and Management: Posts on management technical debt, leading through change, metrics that actually matter, and what I wish I’d known about management earlier. These came from real experiences building teams and making the mistakes you only learn from doing.

AI and MCP: Extensive coverage of the Model Context Protocol ecosystem, from prompt injection governance to structured outputs to building MCP servers. Also practical posts on LLM prompt caching, structured extraction frameworks, and when to use native APIs versus validation libraries.

Python Development: Repuprposed many of my own notes and learnings to do a pretty extensive series on topics related to writing and maintaining python libraries.

Venture Capital: Writing about integrated venture engineering, competitive intelligence through job postings, and how to read job postings as a candidate versus hiring manager versus HR partner.

The Atlanta AI Dinners

This year I started hosting monthly dinners for folks working on AI in Atlanta. Small groups, interesting conversations, no agenda beyond connecting people doing interesting work. In total we did 7 events, ranging from just a few people to a dozen, and each one had it’s own spark.

My goal for this personally was to meet one new interesting person per month, and we smashed that goal, it’s been a delight.

What Worked

Writing regularly: Publishing consistently, built momentum. The discipline of shipping matters more than perfection. Most posts were written a month or more before being published. I would sit down when I had some time and write a full series, or do just outlines for all the ideas I had, or go through and edit drafts in bulk. Doing a batch process like this was more maintainable for me at least.

Personal software philosophy: Building tools for myself first, worrying about polish later. Both ContextSwitch and Evergreen work well because they solve my exact problems without compromise. Building things for me helped me learn new concepts faster and write as both a user and a developer.

In Depth Series: Taking a concept or an idea (“Can we learn something about agent systems from decision theory?”) and exploding it into dozens of posts was a really useful way for me to learn and it seems like others enjoyed reading along.

Looking Ahead to 2026

In 2026 I hope to continue going on increasingly fun adventures with my wife and daughter, eating well, writing intersting software, learning intersting things, riding great trails, and pushing the limit of what we can do in VC with data.

I hope you’ll join me.

Here’s to 2026.

• Will