Friday miscellany (software helpers, this week in screenshots, Texas)

An attempt to collect scattered thoughts (and lower the bar to posting)

It’s Friday. I often jot down thoughts throughout the week, but rarely get to flushing them out. Over time, this creates a graveyard of lost but probably useful provocations and musings.

So, some half-formed lines of thought.

(Also, at bottom! “This week in screenshots”)

Texas eligibility staff are whistleblowing!

Fascinating!

Inside is a link to one of the whistleblowing emails. It’s here. Here’s the beginning:

Let me just say — it is surprising to me that concernedtexans@yahoo.com was available. Are Texans just, generally, not all that concerned? Or did they try Gmail first, and Yahoo was their backup?

(This is in jest; have to laugh as you cry reading some of this.)

Software helpers (“agents”) that interact with government systems

(Or, why is everyone having to repeat the “TurboTax for X” idea maze?)

I spent much of this summer lucky to take part in a residency with a number of incredibly interesting folks pursuing ideas that are, I would say, non-obvious in the how.

One of the topics I thought about most was software helpers (agents) that navigate government services for people.

These things pop up everywhere. And yet… it feels like no one has worked out the core (norm-based) infrastructure for these things to just be normal, trustable, and beneficial?

Scattered observations:

• The truth if you can do these things well or poorly. There are good, bad, and mixed examples all around. Ultimately it’s probably about incentives at the end of the day.

• We have lots of non-software helpers for government already. Tax prep volunteers, SNAP assisters at food banks, health care brokers/navigators, legal aid, social workers. We have lots of these already. And…

• Non-software helpers often have existing legal structures to use. For example, SNAP (sorry to be that guy, but it’s what I know most deeply) has the authorized representative (AR) structure (generally used in more extreme circumstances) but also something many community-based assisters use, a lighter-weight “release of information” (ROI) which is relatively standardized as a form and with policies around it. ROI isn’t as far as I know codified, but it is normal, i.e. accepted.

  • There is also a software analogy here! ROI is effectively read-only (you can get info, but can’t take actions for someone), whereas AR is read-and-write (can do both.)

• There are also some existing software helper structures for government — but they are unevenly spread across programs.

  • Obviously tax filing has a whole ecosystem of 3rd parties who integrate directly with the government systems, and effectively act on people’s behalf.

  • Health care enrollment has brokers, but also some tech-first enrollment services. Specifically something called “enhanced direct enrollment” is worth thinking about. From CMS.gov:

    

• Incentives could be designed for these, and start to approach pay for performance? Think about it this way: we know IRS Free File has poorly designed incentives. TurboTax hid their free file option because it was profitable under the policy design. But let’s think of an alternative design:

  • Imagine 3rd party software helpers were paid by the government (maybe just for lower-income users, maybe for all) on the basis of

    • (Volume of users served)

    • x (Some quality/accuracy measure)

    • x (Some standard customer service metric)

  • Pair this with some annual auditing, and it becomes interesting! All of a sudden, a software-based helper that serves a niche community might pen out. Relatedly…

• We probably should be thinking about this now — new AI probably makes software helpers on government much cheaper to build, and (I suspect) they WILL be built whether we plan for it or not.

  • So what might planning effectively be? Maybe it’s…

    • Creating some industry norms/standards (“we are Good Software Helpers™ because we commit to…“) that 3rd parties building these things agree to (e.g. no overnight sudden user volume increases, a security audit, certain required disclosures, including a complaint form that goes to the underlying government agency)

A final philosophical point: a helper is sometimes genuinely serving different interests! Improving the underlying government service and enabling more helpers are not at odds. (In fact, helpers probably de-risk potential changes.)

I know lots of people are building this (and being like “should we ask them for an API????”)

But who is talking about this, at the norms and governance level, above Selenium web driving scripts?

This week in screenshots

A Friday thing. Select screenshots from my week. No, no context provided.