Hammock driven development

You’ve heard of TDD, and more recently also SDD (spec driven development)… but have you heard of HDD — aka Hammock driven development?

I recently came across a video gifted to me by the YouTube algorithmic gods with a catchy enough title that sounded more like a “honest bait” than a clickbait, it was titled: “Hammock driven development”, talking about an alternate approach to software development. I’ve currently been thinking a lot about various new processes that could improve software development, and this lecture from Rich Hickey from 13 years ago seems like one of those older ideas that need a new revival story. Especially now.

So I jumped right in.

I try here to write in my own words what I understood from the lecture and not refer to any of the supporting transcripts or Youtube timestamps (it’s now ‘fresh’ in my mind and want to make use of this moment in time)

---

We have two types of minds: the waking mind, and the background mind. Historically we’re quite used to the system 1 and system 2 framing by Daniel Kahnemann, but this is far more encompassing than I’d expected, the background mind here, the OP refers to being generally good at strategic, holistic thinking.

You would normally want to leverage such strategic tasks for the background mind. Not that the waking mind doesn’t do strategy, but it’s more focussed on the input <> output processing. And as such its results are much more strategic and immediate output oriented. The OP then suggests to leverage this partisanship to our interests.

While writing code, we are thinking through problems and we should first know how to draft a problem, and this could be in terms of scope, constraints, framing etc, we need to get that right first, and then when we have to start solving problems, at times we might encounter harder problems which were not used to encountering before (not the usual fetch from a dB and display it on the CRUD UI types).

When this happens, we then need to think and segregate such problems into the known types which can be easily done by the ‘waking mind’, and the harder subset which needs to be delegated to the background mind. The background mind works in interesting ways, you would not know in advance what the solution might be, you just need to assign it to this background mind and see what the ‘eye of the mind’ unravels.

But you will get there nevertheless, and to let this slow burn process happen without any stress it needs. In fact stress environments make you go into the ‘waking mind’ mode, and you would not be able to do the slow burn. That’s also one of the reasons why this should be done through a shower thought.

OP also recommends Michael Poyalyi’s How to solve problems book as it gives a much more mathematical rigour to the varied approaches to solve problems. There is another book I was able to find on the YT comments called ‘How to solve with computer’ that gives an algorithmic perspective to solve problems. This is another one on my Umberto Eco’s antilibrary style to-read list.

This time when I listened to the lecture, it hit me in a different way especially since I’ve also been fascinated about the AFK (away from keyboard) and HITL (human in the loop) segregation of work by Matt Pocock, an AI tutor who recently shared this in his talk at the AI engineers forum.

The AFK mode is usually for known solved problems, it’s similar to the waking mind for the agents, where they could run autonomously and completely solve the problem/s without causing much tech debt.

The HITL mode is usually for problems that require a feedback-loop type dynamic with humans to provide the right inputs for it to solve the problems better. This works well for human guidance aided by the right ontology. In the HITL mode, for example in case of 1 and 2:

1:

“I have a deep pimple only on my chin is it more likely from diet or hormones”

2:

“sudden cystic acne isolated to chin, is the more likely etiology hormonal fluctuation or pro-inflammatory gut disruption from dietary changes?

Both 1 and 2 are effectively the same question but 2 provides you vastly different (and better) responses. There is also a Harvard AI safety study which talks about this method, where changing ontology provides different responses.

And for AFK mode, you might not even need such thoroughly grounded ontology. It gets the aim, the objective and the steering right in one go.

Now coming back to the original topic by OP about the waking mind and background mind, and also finding similar analogies in the agentic coding world with the AFK and HITL types, I do think there is a bigger set of problem space where despite the best of agentic coding modes, both HITL and AFK, you would need to consciously use your background mind, for days, weeks or months to probably crack the code.

This process might take days, or months. A question I have for the reader here is how rare or common has it been for you to be in this background mind mode trying to solve a hard problem? Me, personally, it’s been quite rare. It’s such a terrific experience to be in this mode of problem solving, where you don’t want to be disturbed, and neither do you should be in front of the computer.

You should be lazying around, doing random chores, or as the OP suggests, if you do have a hammock in your background, then that’s all you need.