AI, Dopamine, and the Night I Rewired My Home Lab⌗

A few nights ago, I opened my real backlog.

Not Jira. Not a curated roadmap.

The actual list:

stabilize my k3s cluster running across Raspberry Pis and an old desktop
build a network sentinel agent on my OpenWrt router to detect malicious traffic
improve observability (Prometheus, exporters, Grafana dashboards)
monitor smart plugs for power anomalies
orchestrate domotic scenarios based on real signals (not just timers)

This is the kind of list that usually grows.

Not shrinks.

Then I turned on AI⌗

I had AI wired directly into my environment.

And instead of prioritizing…

I just started executing.

One task.
Then another.
Then another.

Bang. Bang. Bang.

Exporters deployed
Metrics flowing
Alerts firing
Dashboards shaping up
Agents scanning traffic

At some point, I wasn’t implementing ideas anymore.

I was discovering systems.

Security changes when exploration becomes cheap⌗

Here’s what surprised me the most.

As I instrumented the system:

smart plugs started exposing power patterns
network flows revealed unexpected behaviors
devices that “looked fine” showed anomalies

Not vulnerabilities in the traditional sense.

But weak signals:

unusual consumption patterns
unexpected network chatter
implicit dependencies between devices

Things I would have never had time to explore manually.

And that’s when it clicked:

With AI, security is no longer just about known threats —
it’s about exploring unknown states at scale

The dopamine spike⌗

Somewhere in the middle of all this, I paused.

Because I felt it.

hyper-focused
highly motivated
unable to stop
constantly jumping to the next improvement

It felt like flow.

But more intense.

And I asked myself:

Is this just productivity… or something else?

Engineers + AI = infinite execution loop⌗

If you’re in infra or security, you already have the mindset:

curiosity-driven
system-oriented
always seeing improvements

Normally, execution is the bottleneck.

AI removes that.

Now:

writing exporters takes minutes
building agents is trivial
testing hypotheses is cheap
iterating is almost free

So what happens?

You stop choosing carefully — and start executing continuously.

The loop I fell into⌗

Identify a gap (monitoring, security, automation)
Build a quick solution with AI
Discover new signals or anomalies
Expand scope
Move to the next idea

Repeat.

This feels like peak productivity.

But it’s actually:

A dopamine-driven infra loop

Why this is different from other “dopamine traps”⌗

We’ve seen loops like this before:

social media
gaming
dashboards for the sake of dashboards

But this is not consumption.

This is:

dopamine attached to building systems

And that’s why it’s dangerous.

Because it looks like progress.

The real risk in infra & security⌗

The risk is not addiction.

The risk is:

building systems that are wide… but not deep

You end up with:

10 dashboards, none fully reliable
multiple agents, none production-ready
partial observability
fragmented automation

In security, this is worse than doing nothing.

Because:

false confidence is more dangerous than no visibility

What actually worked (lesson learned)⌗

After that session, I changed one rule:

Build one thing. Then make it real.

Not:

“good enough”
not “it works on my machine”

But:

observable
reliable
documented
explainable

Example: from toy to system⌗

❌ Before (dopamine mode)⌗

exporter deployed
metrics visible
quick dashboard
move on

✅ After (controlled mode)⌗

exporter deployed
metrics validated (correctness > existence)
alerts defined with real thresholds
failure modes tested
dashboard tied to action
integrated into a workflow (not just visualization)

Smart plugs: a concrete case⌗

Monitoring smart plugs started as a small idea.

It became a system.

From:

“see power consumption in Grafana”

To:

detect abnormal usage patterns
correlate with device state
trigger domotic actions:
- shut down unstable devices
- alert on unexpected consumption
- adapt behavior based on load

This is where infra meets automation.

And where:

observability becomes control

OpenWrt agent: another example⌗

The idea:

detect malicious traffic on my home network

With AI, it was easy to:

parse flows
classify patterns
generate alerts

But the real work was:

defining what “malicious” actually means
reducing false positives
integrating with existing signals
deciding what action to take

That’s the difference between:

a demo
and
a security system

AI is a cognitive amplifier (and a stimulant)⌗

The closest analogy I’ve found:

AI behaves like a cognitive stimulant.

Not chemically.

But functionally:

faster feedback
higher engagement
reduced friction
continuous reward

Which means:

you need discipline at a different layer

A better way to use AI in infra & security⌗

Here’s the model I’m converging on.

1. Exploration is allowed — but bounded⌗

Use AI to:

explore ideas
generate prototypes
map solution space

But set limits:

time-box it
define exit criteria

2. Execution is sacred⌗

For anything that matters:

validate signals
test failure modes
define ownership
connect to action

If it doesn’t change behavior:

it’s noise

3. Observability → Decision → Action⌗

Every system you build should follow:

Observe (metrics, logs, signals)
Decide (rules, models, thresholds)
Act (automation, alerts, orchestration)

If you stop at step 1:

you built a dashboard, not a system

4. Build less, but build deeper⌗

AI makes building easy.

Value still comes from:

reliability
clarity
integration

Final thought⌗

That night felt like unlocking a new capability.

Not because I became better.

But because:

the cost of turning ideas into systems collapsed

And when that happens, a new problem appears:

choosing what not to build

If you’re in infra or security, this matters.

Because the goal is not to build more systems.

It’s to build:

fewer systems that you can actually trust

Build fast.
But finish one.