I'm in Michigan again, after being engrossed in the proceedings
of the 2026 Code for America Summit in Chicago.

The main lesson? Where government meets technology,
there is only room for CEOs to speak up.

- - -

Here's some background on my relationship with the organization;
copied from my recent cover letter to them in an application from January.

> I remember the long run of Wednesday nights
> that I spent at CfA headquarters in SoMa,
> before I ever applied to be a fellow,
> pushing along on the program at
> [Code for San Francisco]("https://codeforsanfrancisco.netlify.app/")
> (now called [SF Civic Tech]("https://www.sfcivictech.org/")).
> The next year, as a fellow in Seattle,
> we had a chance to speak to [Open Seattle]("https://openseattle.org/").
> Years on, I checked in on [Code for Philly]("https://codeforphilly.org/").
> They turned me away that night,
> because I had forgotten to bring my vaccination card.
> Ironically, by then my vaccination card was being signed
> by a medical officer in the House of Representatives.

> Zoom into the middle of that span, though -
> and during the pandemic I also honed in
> on [Chi Hack Night]("https://www.youtube.com/@ChiHackNight"),
> and [Lansing Codes]("https://www.lansing.codes/").
> This was a challenging moment for all of us,
> because the organization where we had found a home
> seemed to be closing the doors on its most public forum, the brigades.
> [I had one good idea of what could happen]("https://discourse.codeforamerica.org/t/announcing-the-network-revisioning-project/1058/6?u=c4lliope"),
> though none of us had any idea of all that would.

More than any other group, Code for America has seemed like a philosophical approach;
their name appears to command, or demand, the action they seek.
Now, rather than being read as an order,
their name displays as a menu, only a single option:
"Code, for America, from Anthropic AI. Constitution included."

Code for America announced their recent partnership with Claude;
and I am saddened beyond measure to see the consequences clear as day.

Code for America handles the procurement issues
for municipal, state, and federal agencies
to procure access to the AI they seek
as a guarantee for early, phased-in retirement.
Now, the AI that Anthropic has been selling is going to be packaged cleanly,
in vibe-coded and tailored-to-size apps for one purpose or another;
from food aid to automated evictions.
"The strategy is delivery", after all,
in a nation whose primary product is munitions.

- - -

Sadly, this could be flipped around - or rather inside out - so easily.
Bring an engineer rather than a CEO onto the stage,
and we may be reaching through to the demands and clear direction
that solidarity groups have been promoting for decades, normally unheard:
open access to the means of production.

In this hype cycle, AI is the product;
the means of production is access to the underlying, open-source mechanisms:

- the vector databases,
- the encoding models,
- the decoding models,
- the tokenization models,
- the similarity search engines,
- the graph databases for mapping knowledge,

the many pieces whose "sum of the parts" is called AI,
though which each in measured additions or absence in a product
form the desired recipe that is called for on one occasion or another.
An example? You can make more humane decisions by handing a case-manager
direct access to a knowledge graph (basically imagine Notion or AirTable),
rather than obscuring the facts of the case
through an encoder (email correspondence -> assumed facts),
and a decoder (recorded facts -> assumed conclusion and irrevocable consequence).

- - -

The one session I did find some solace in though,
was the discussion of the group collaborating with IRS direct file;
they open-sourced their "fact graph" of dependencies for IRS tax return forms.
For more research and open-source programming along those lines,
see [PolicyEngine](https://www.policyengine.org/us).

I made a small proposal to a colleague in the first breakout room of the conference,
to examine how often we would hear the phrase "AI" compared to "vector database";
the concluding ratio was as predicted, many-to-zero.

These CEOs charging headlong into the camera lens on stage
know that their product is made of a delicate balance of internal mechanisms,
that a small shuffle from one direction to another, a dialing in of temperature,
can produce the difference between a swell of emergency housing response
and a cascade of evictions when people become a problem.
The logic of resettlement can be run from a single dial:
"Population from 11 to 3, please."

This experiment has played out before,
during the election cycle for John F Kennedy,
in a corporation called [Simulmatics](https://en.wikipedia.org/wiki/Simulmatics_Corporation).
Simulmatics built a "People Machine" for the 1952 presidential election,
which predicted the electoral consequence of polling inputs.

Our modern "People Machine" goes much deeper than polling numbers -
it has a full model of the language
used to relate and grasp hold of our emotions.
How real can this "People Machine" be when making humane decisions,
and how shell-shocking the consequences when they fail to comprehend humanely?