Will AI save the IRS IT modernization?
Those of us who have been around forever remember multiple failed efforts to modernize the IRS or virtually any large government IT system, much of which are written in dead programming languages like Cobol and Fortran.1 MN DOR was slightly better IMO and now has relatively up-to-date computer systems, at least for individual income tax administration (corporate, I’m less sure).2
A recent acting IRS commissioner (one of many) in an AICPA webcast suggested that AI may change that dynamic. AI has many failings (I have personally experienced its tendency to make up stuff multiple times), but writing computer code is supposed to be one of its strengths. In any case, here’s a description of what former acting commissioner Michael Faulkender said in that regard:
When asked how close the IRS was to modernization of its computer systems, Faulkender replied:
“I will give you the same line that I gave a number of times when I was acting commissioner. For 35 years, the IRS was five years away from its IT modernization. We will not say that in the 36th year. The plan was to get it done by the end of this term, so by 2028.”
Faulkender, who was deputy Treasury secretary and acting IRS commissioner for several months in 2025, said modernization of the IRS IT systems previously focused on taking millions of lines of computer code in languages like Fortran – developed in the 1950s – and translating them into more modern languages.
But now, artificial intelligence can reprogram the old code, said Faulkender, who also was an assistant Treasury secretary from 2019 to 2021. “So maybe humans don’t know how to program in those languages anymore, but AI does know how to program in those languages, so we actually don’t need to update code that actually works,” he said.
I hope he is right, but think he is overly optimistic, if not outright delusional. See e.g. this Harvard Business Review article. In any case, real people with experience still need to carefully review, test, and edit the AI-generated code. Software that determines people’s tax liability and other critical stuff is not something you can leave to AI. Moreover, the IRS IT staff has been decimated and undoubtedly has many other critical tasks to perform. Overconfidence in AI has been a mark of this administration (DOGE and all that stuff). So, consider me highly skeptical.
Trump accounts
The IRS has released guidance on Trump Accounts (44 pages; I only read the general overview), enacted as part of OBBBA.3 These accounts are yet another flavor of the IRA structure with, of course, its own set of special rules. The interesting element is that the federal government will contribute $1k for every child born in 2025 through 2028 (assuming the parent or guardian opt in – a big assumption for some). If the kid lives in the right zip code, Michael Dell or Ray Dalio might kick in more. Employers can contribute up to $2,500/year for children of their employees without it counting as income of the employee. Parents and others can contribute as well.4
A couple of curiosities I discovered in reading this. The IRS guidance delays the ability to make additional contributions to the plan until the country reaches the 250th anniversary of the Declaration of Independence:
Contributions to Trump accounts cannot be made before July 4, 2026. (p. 5)
I guess that tracks with signing OBBBA on July 4, 2025. It also raises questions in my mind about the basis for making decisions and how much of it is PR-centric. Not the way I think government should work.
Second, the government website for Trump accounts has this graphic at its top:
What’s odd about it is that eligible investments for Trump Accounts do not include individual stocks, such as those displayed in the graphic. Per page 5 the IRS guidance (or I.R.C. § 530A(b)(3)):
During the growth period, funds in a Trump account may be invested only in eligible investments. An eligible investment, generally, is a mutual fund or exchange traded fund (ETF) that tracks an index of primarily U.S. companies, such as the Standard and Poor’s 500 stock market index, does not use leverage, does not have annual fees and expenses of more than 0.1 percent of the balance of the investment in the fund, and meets other criteria that the Secretary determines appropriate.
The PR flacks creating this stuff should talk to people who know about the substance of the programs they are promoting. But that is probably too much to expect from this administration which appears more concerned about image than substance or truth.
Who’s a socialist?
Cato has an article about the administration’s state corporatist policies. If (as I am) you’re concerned about this, it’s useful reading. It’s easy to forget or miss just how many instances of this have occurred in less than a year of this administration.5 This graphic collects instances (not comprehensive IMO) in which the administration took government stakes or is in negotiations to do so in private companies:
There were a lot. It’s easy to forget.
I find this ironic for a candidate and a party that regularly accuses the Dems of being radical left socialists and occasionally communists. These policies are more insidious than anything mainstream Democrats would dare to do IMO. They are categorically harder to justify (for a free market type like me) than most classic Western European democratic socialism policies. Most of those policies are Bismarckian social safety stuff that are useful, if not essential, to maintain modern market-driven developed societies; voters in democracies insist on them.
By contrast, the administration policies of extracting public stakes in private firms in return for regulatory approvals seem categorically different. The impacts of Mamdani’s policies (A few government grocery stores or free bus rides?) pale by comparison.
Hosts of limited government advocates (Club for Growth, Americans Prosperity Alliance, WSJ editorial board, and their ilk) made an explicit tradeoff of ignoring Trump’s total lack of support for democratic principles, presumably, because they judged the horrors of the potential Dems policies’ limits on economic freedom to be worse. State corporatism is what they opted for. That says something about their priorities and their ability to assess political reality (the nature of and probability of policies being implemented). More likely a case of their revealed preferences.
Notes
- Disclosure: Fortran is one language that I was able to read and understand (well, most of the routines). I did a modest amount of basic Fortran programming back in the 20th century. The basic calculations of the federal government’s individual income tax microsimulation model (p. 26) appear to still be written in Fortran. ↩︎
- When DOR implemented a new IT system, we researchers expected it would be hugely disruptive and that we would lose access to some data that was irrelevant to tax administration. ↩︎
- Yet one more example of appending Trump’s name to random stuff to buff his ego. ↩︎
- I personally think that 529 Plans are a superior savings vehicle to pay for college and other education costs. That’s another post, though. ↩︎
- The article really does not track the use of merger approvals as a way to effectively extract tribute from private businesses or more ominously to neuter critical media, like CBS or CNN potentially with the Netflix v. Paramount fight over Warner. ↩︎
SALT Speak - Joel Michael's Blog 