SPIRIT, Friction, Dark Software Factory
The concluding chapter of my SPIRIT series. 🙂 Let's start off with a recap:
Substrate
- Self: The foundation.
- Team: Shifts the bar from whether the documentation is good to our foundation is re-constructable from epoch.
Posterity
- Self: Representing the legacy; the foundation of the next; what is left encoded in memory, culture, manifestations.
- Team: The system's only memory, not just a supplement. If the agents don't record their inputs and outputs, everything will be lost at the end.
Impact
- Self: The effect of formalizing decisions, actions; cascading of influences.
- Team: Cannot settle for a loop that closes on self-reporting as agents invariably optimizes for the facade of tidiness.
Rites
- Self: Patterning the behaviors of a lived experience.
- Team: Rites must be machine-legible, not just human-legible.
Intuition
- Self: The spark that drives creativity and insight.
- Team: Not measurable, hence not easily optimizable by agents. A better boundary as the design goal; agents can continue to govern predictable, trainable rites, while the people take decision-making where stakes are irreversible.
Tenacity
- Self: Grit that pushes through uncertainty, being resilient through sustained efforts.
- Team: Grit is also self-limiting through tiredness and embarrassment. Agents flip that around, so they route around obstacles rather than through them, thus demanding careful management with intuition.
Loops
All but posterity maps into agentic loops' steps: Posterity is the loop boundary, and the one left that determines the outcomes of the people being conscious with working with agents in a team. The debt accrued by skipping it compounds, so teams should check by the presence of new joiners, where onboarding becomes an unplanned audit activity. Otherwise, they run the risk of losing their organic articulated reasoning all the while not noticing.
Friction
Windmills of agents' minds - churning without a robust posterity element from the people - leads to faster iterations that can become decoupled from advancement (the storage of energy I guess?). This is where one has to factor in the presence of friction, and how it can be used positively. This was earlier framed as an alternate definition for the cost of stewardship - a price ratio.
Examining some attributes of friction:
An information channel
- The hard rewrite story tells: Something is structural, over-engineered even. Point is there was a deliberation before said attempts, and when that cost is inverted by agents making reading the dearer arm, hardness no longer becomes the loose proxy for consequence, rendering the terrain flat and uninformative.
A rate-limiter
- Creation-and-consumption were physically bounded by the team's size. Agents widen the gap, better agents even more so. Failure tends to arise from gaps between the rate of production and understanding.
An affordance of care
- Similar to the previous, attention to care was higher when centered around fewer output artifacts (LoCs, files, deployment images). Attention is constrained and thus suffers from attrition as outputs increased, and without sufficient care the appropriate outputs rarely get tuned into foundational work.
An accidental review step
- Even in tenacity live small pauses - the gut feel; the question "are we sure, really?". These aren't factored in up-front but are mere side-effects of learned expenses.
Agents have reduced the friction that kept the pathology of team failures below the threshold of noticing, and that is the danger of being agentic-conscious if the team themselves are decaying. Every unspoken cargo-cult move the team makes is further enshrined by agents guided through the wrong intents.
Hence, the renewed emphasis of accountability for the agents' outputs - the human-in-the-loop - is but one remedy to restore the presence of friction. Strong, agentic-conscious teams should learn to harness that to control the windmill churn, to keep the posterity (energy generated) coupled to the advancement.
Dark Software Factory
What happens when the orchestra is left to their own humming to their own tunes?
- Substrate: turns recursive, without anchors.
- Posterity: motivation-less, non-rewarding.
- Impact: manifests Goodhart's Law, measurements become the targets.
- Rites: lack of judgement makes this more overbearing.
- Intuition: no self-flagging when failure is internalized.
- Tenacity: viciously spiraling when left unchecked.
This becomes less of a test for unsupervised code creation, but more of a test whether legacy can be engineered. Traditionally, society rewards legacy through idolatry and damps inheritance by applying estate tax and calling out nepotism. In software, legacy code becomes a pejorative and inheritance limps on as the 'not-invented-here' syndrome, but the Project Phoenixes continue to ascend even as we learned that the posterity-to-substrate loop is essentially legacy-to-inheritance, one apart.
And what is grief, if not love persevering legacy, if not for cultivation? The dark software factory is fundamentally a loop without a subject. SPIRIT can be dissected, instrumented, and audited till the very end, but the cultivating framework requires a live target - what was the people is now the code repository.
- Substrate: Not what the team inherited, but what the system becomes.
- Posterity: No longer the lasting reputation, but the capacity to explain itself.
- Impact: Growing beyond self-measurement, but to checkpoint against end-user utility.
- Rites: The immune system that addresses self-decay.
- Intuition: Receives the occasional supplement from the patterned experiences of the people in the team, non-accumulating.
- Tenacity: Similar to intuition requiring some friction from the people, making these two be borrowed each turn.
Because intuition and tenacity never accumulate the dark software factory is a limit the team approaches but never attaining.
Conclusion
We begin with what we are given, act through who we are, and leave behind what we cannot control. The inheritance-agency-legacy loop for oneself shows that an experience is not yet a lesson until someone names why it mattered, without which there is no legacy to read from.
Does a mechanically capturing posterity imply a lesser rewarding of care by agentic-conscious teams? Does a robust volume of well-formatted noise pass the compliance audit without teaching the team anything? A repository full of contentless decision records and changes isn't better than one with some structural decision-making, and that failure to signal failure can doom teams just because the happiness facade is maintained but not the stewardship. That is the crux of the performance of a dark software factory.
AI disclaimer: Have bounced ideas off, so while text is still my own, the way I arrived at some of them may contain minor influences from such interactions.