This example elegantly shows how a few dozen lines of Clojure can replace an entire accounting application. The transactions live directly in the code, the categorization rules are simple pattern matchers, and the "interface" is just printed output of the transformed data. No SQL, no UI framework, no MVC architecture - yet it solves the actual problem perfectly.
The power comes from removing indirection. In a conventional app, you have: - Data model (to represent the domain) - Storage layer (to persist the model) - Business logic (to manipulate the model) - UI (to visualize and interact)
Each boundary introduces translation costs, impedance mismatches, and maintenance burden.
In the LISP approach shown here, those boundaries disappear. The representation is the storage is the computation is the interface. And that direct connection to your problem is surprisingly empowering - it's why REPLs and notebooks have become so important for data work.
Of course, there are tradeoffs. This works beautifully for personal tools and small-team scenarios. It doesn't scale to massive collaborative systems where you need rigid interfaces between components. But I suspect many of us are solving problems that don't actually need that complexity.
I'm reminded of Greenspun's Tenth Rule: "Any sufficiently complicated program contains an ad hoc, informally-specified, bug-ridden, slow implementation of half of Common Lisp." The irony is that by embracing LISP principles directly, you can often avoid building those complicated programs in the first place.
Rich Hickey's "Simple Made Easy" talk explores this distinction perfectly - what the industry calls "easy" (familiar tools and patterns) often creates accidental complexity. The approach shown here prioritizes simplicity over easiness, and the result speaks for itself.
In toy programs, such as this one, those concerns can all be discarded, so LISP is a good fit. But in serious systems it’s soon discovered that what is offered is in fact “simplistic made easy”. That’s not to say that traditional systems don’t suffer from all the ills Hickey diagnoses in them, but that we differ on what the cure is.
Here's the issue. Starting out you almost certainly don't need that rigid interface. However, the longer the app grows the more that interface starts to matter and the more costly retrofitting it becomes.
The company I currently worked at started out with a "just get it done" approach which lead to things like any app reaching into any database directly just to get what it needs. That has created a large maintenance issue that to this day we are still trying to deal with. Modifying the legacy database schema in any way takes multiple months of effort due to how it might break the 20 systems that reach into it.
When the system is small, it makes a great deal of sense to be an artisan and design simple automations that work for exactly that task, which for the most common things is always supported by any production-oriented programming environment - there's a lot of ways in which you can't go wrong because the problem is so small relative to the tools that any approach will crush it. "Just get it done" works because no consequence is felt, and on the time scale of "most businesses fail within five years", it might never be.
When it's large, everyone would prefer to defer to a common format and standard tools. The problems are now complex, have to be discussed and handled by many people, they need documentation and clear boundaries on roles and responsibilities. But common formats and standards are a pyramid of scope creep - eventually it has to support everyone - and along the way, monopolistic organizations vie for control over it in hopes of selling the shovels and pickaxes for the next gold rush. So we end up with a lot of ugly compatibility issues.
In effect, the industry is always on this treadmill of hacking together a simple thing, blowing out the complexity, then picking up the pieces and reassembling them into another, slightly cleaner iteration.
Maintenance can be done successfully - there are always examples of teams and organizations that succeed - but like with a lot of infrastructure, there's an investment bias towards new builds.
The whole post was about how that doesn't solve the problem perfectly - there is no way to interactively edit the output.
> by embracing LISP principles directly
This could just as easily have been javascript+json or erlang+bert. There's no lisp magic. The core idea in the post was just finding a way for code to edit it's own constants so that I don't need a separate datastore.
Eventually I couldn't get this working the way I wanted with clojure and I had to write a simple language from scratch to embed provenance in values - https://news.ycombinator.com/item?id=43303314.
Spreadsheets. If you squint the right way, they embody the lisp principles for the non-programmers' world.
No SQL but in it's place is some code. The point of SQL was to standardize a language for querying data. This is just using a language other than the standard. A UI is a way for people to avoid writing code.
Sure doing your own custom thing results in something easy for the programmer. Nothing new about that.
I worked with somebody with your name in the early 90s on a Sequent/Dynix system, that wasn't you by chance was it?
1. Write queries in datalog
2. Compose them like reusable functions
3. Stream that data between SQL db and dataframe columns and graph db
4. Both also supported with reusable composable function interfaces
5. With IDE/notebook feedback and content-assist including not just syntax but the referenced data model.
6. With distinct modes for exploring datasets and for generating code and tests for hardened pipelines.
7. With abstract accounting for operations of the db, transforms, and transfers.
Data and language will always have ultra-specialized forms. I’m looking for a low-overhead way to explore solutions using different combinations of baseline paradigms before generating code for the one to productize.
https://www.scattered-thoughts.net/log/0027#preimp
https://x.com/sc13ts/status/1564759255198351360/video/1
I never wrote up most of that work. I still like the ideas though.
Also if I had ever finished editing this I wouldn't have buried the lede quite so much.
I belive it works for you, and I can even say I think most projects have (a little) too few lines of embedded sql in them.
But arguing for code that isn't checked out with the rest of the code when I clone the repo, isn't greppable, and radically departs from everything most people are used to, that isn't something I'd accept.
Complicated formats can't be meaningfully version controlled.
Excel guy: Boom got it done in 5 minutes sips coffee what's next?
10x engineer: I have devised a new programming paradigm to tell me I need to spend less on coffee, wrote a blog post about it and missed a call from a client and a recruiter.
While I was working on that, the company was busy trying to figure out which other companies the millions of unaccounted excess dollars in their fiduciary account belonged to.
Turns out the boom-got-it-done-in-5-minutes types don't always have a plan for making the big picture work, beyond their current 5-minute problem.
Non computer people try to solve a problem internally with existing tooling, but the scale gets out of control and they hire a specialist, whether accounting or computer scientist. Most of the jobs I have consist of a company already hacking together a solution from canned software or having a prototype of what they wanted, don't think this is weird at all.
Are you suggesting they should have built a well architected Java program from the start? I'm not so sure.
1 - It's easy to say with monday's newspaper, companies can move fast and then worry about the consequences later if they grow and are succesful. Internal proyects or the whole company can be unsuccessful so you can tag on tech debt and worry about it if it needs to survive, happens all the time. As a curious note, the legal system supports this, companies have limited liability so courts are ok with letting companies that go too deep into the red (tech debt or cash) collapse into 0.
2- Overengineering is also a risk, although usually the failure mode is it never takes off the ground, but what's to say the couldn't have hired a -10x engineer to build their accounts system and then needed to hire an actual 10x engineer to fix the mess?
3- This seems like an accounting problem domain, I think if a company has classically trained accountants you avoid this problem, and if you don't then you get this problem regardless of whether using excel or a programming language. Double entry bookkeeping has been around for centuries. (I'm guessing an accountant helped build the specs for whatever you built for 1M$). Good accountants can handle huge (+1M) companies with pen and paper, and regularly do so with Excel, so I'm not buying that excel was the culprit here.
That said 1M$ is a lot, congrats!
Evaluating the problem tells you how complex the solution needs to be. This problem of categorizing basic financial transactions is pretty basic and I'd take the basic approach every time.
But I think the point here was exploring different paradigms and their trade-offs. I appreciate that a lot, as I think we're stuck in various local optima as an industry - and spreadsheets seem to be one of them (once they pass a certain complexity threshold).
I'm not sold on the "everything in clojure" model here, I think you could accomplish all this does with a script written in $LANG and a little DuckDB (and probably a half dozen other similarly suitable approaches), but again - I appreciate the goal of exploring the solution space, especially when it's using approaches that I'm not that familiar with.
The Excel approach is absolutely more efficient for one-off tasks or where the goal is simply "get numbers, make decision, move on." No argument there.
Where the programmatic approach shines is when:
1. The same task repeats annually/monthly (like the author's accounting example) 2. The rules or categorizations evolve over time 3. You need an audit trail of how you arrived at conclusions 4. The analysis grows more complex over time
I've seen plenty of Excel wizards whose spreadsheets eventually become their own form of programming - complete with complex macros, VBA, and data models that only they understand. At that point, they've just created an ad-hoc program with a different syntax.
There's a sweet spot for each approach. Sometimes the 5-minute Excel solution is exactly right. Other times, spending an hour on a reusable script saves you 10 hours next year. The real 10x move is knowing which tool fits which situation.
And yes, sometimes we programmers do overthink simple problems because playing with new approaches is fun. I'll cop to that!
Excel wasn't built out of a small personal project, it was a huge multi engineer project built on microsoft (built upon decades of spreadsheet software tradition and arguably centuries of accounting tradition), by a company that had a huge userbase that acted as stakeholders for managing the data and accounts of multi million dollar companies and small businesses alike.
I think that complex contraptions arise more from complex necessities than overengineering a solution from a simple problem
But another program with a similar impact -- dBase, the precursor to Microsoft Access, was. Cecil Ratliff wrote it to manage football statistics so he could win his office football pool.
If you're keeping records as files, I think you should just use a million files instead.
1. Easy to version control
2. You can sync on the cloud with less conflicts if each record is isolated from the rest
3. Sounds unique and cool
Lisp is a fantastic tool for accomplishing this, but to me, Access is tailor-made for exactly this problem (and a whole lot more besides, it's a beast). If one follows the trail of the original problem far enough, I think one is likely to end up with an Access-like solution regardless of implementation language.
If this was a joke, I'm sorry for explaining it.
If this was not a joke, I point you to the giant
<h1>!!! DRAFT !!!</h1>