Founder, SeinX Amplified Intelligence Collective
Here’s how it usually goes...
Promise
Automation promises more progress with less effort
Imagine what we could do
with all that spare time.
It starts as a good idea: easy to justify,
with a clear payoff and hardly a downside.
Oh, the system does that now,
you don’t need to check.
Checks come off and people step back,
with the belief that design can foresee everything,
and the work needs little human input.
Honestly? I’m not sure
how that part even works anymore.
Without the everyday work,
context and skills fade,
and few still see how it all fits together.
Huh, that’s not in the rules.
Anyone know what we do now?
Edge cases resist rules,
each harder than the last,
and fewer people are left to judge them.
It took the automation a second.
Cleaning up is taking us till Friday.
Failures last longer
and recovery costs more.
The time saved goes into repair,
and customers notice first.
We’ve come too far to stop now.
The next version will fix it.
Wins invite more automation,
and failures seem to demand it,
so each round adds more to maintain.
Why does this happen?
Automation goes wrong when
it aims to replace judgement.
Each failure makes going back harder,
because the everyday practice has faded,
and more automation looks like the only way forward.
But each win makes the next round easier to justify.
Going back to doing it all by hand doesn't scale either.
So what's the alternative?
Automation that makes the people
using it more capable, not less.Here at SeinX, we call it
Amplified Intelligence.
Sometimes the answer is modern AI, sometimes a simpler tool, and more often than not it is taking control over the automation that is not serving you.
Time and again, teams that take their automation under control and put people in charge
get quality back.
-
Sendcloud Addresses Complexity cut 2023–2024 Address logic built around Dutch conventions caused trouble during international expansion. A team with authority across domains replaced it end to end and moved traffic across gradually. Parcels needing a human to fix an address fell from about 1 in 10 to about 1 in 100 - Klarna Support chat Replacement reversed 2024–2025 An AI assistant took over two-thirds of support chats, sold as the work of 700 agents. Quality slipped, and Klarna began hiring human agents again. AI keeps the routine two-thirds, people handle the rest, and a human is always available on request
- Airbnb Test migration Automation that amplifies 2024 An LLM migrated 3,500 test files to a new framework, each file passing validation at every stage. Engineers finished the last 3% by hand. 97% migrated automatically and the rest started from working drafts: 6 weeks against an 18-month estimate
- 37signals Cloud exit Cost cut 2023–2024 Moved off the cloud onto its own servers, with the same team running them. Cloud bill down from about $3.2 million to $1.3 million a year, with the hardware paid back within 2023
- Sendcloud Error messages Automation that amplifies 2024–2025 Cryptic carrier errors became short, local-language messages. While AI is used for mapping, the results are reviewed by people. Merchants could act on the message, and support hours for this category dropped 5x
- GitHub Flaky tests Noise pruned 2021 Detected flaky tests automatically, kept builds green and assigned each to its likely author. Flaky builds fell from 1 in 11 commits to 1 in 200, so red means real again
- Segment Destinations Complexity cut 2018 Over 140 microservices, one per destination, had become slow to change. The team merged them into one service. Shared-library improvements rose from 32 to 46 in a year, and one engineer deploys it in minutes
- Prime Video Stream monitor Complexity cut 2023 A stream-quality monitor built from serverless parts hit its limit at 5% of the planned load. The team rebuilt it as one application. Now checks thousands of streams, at about 90% lower cost
- Sendcloud Pricing Automation that amplifies 2021–2025 Prices lived in database migrations only engineers could change. A shared model, exchange format, and robust validation opened them to other teams. Millions more price lines, managed by the same team for years with minimal issues
- Ng Teng Fong Hospital Alerts Noise pruned 2019–2021 The hospital cut its record system’s pop-up alerts by 59.6%. Alerts acted on rose from 8% to 54.7%, with no serious events attributed
- Slack Deploy safety Automation that amplifies 2024–2025 73% of customer-facing incidents came from Slack’s own deploys. Automatic rollback now restores service first, and engineers review the decision afterwards. Customer impact hours from change-triggered incidents fell 90% between early 2024 and January 2025
- Discord Message storage Complexity cut 2022–2023 Replaced a 177-node Cassandra cluster with 72 ScyllaDB nodes. Slow reads dropped from 40 to 125 ms to a steady 15 ms, and the weekend firefights stopped
How we tell the difference
Automation amplifies people when it is:
Continuous
Reviewed, practised and improved as the work changes, so skills keep growing with the system.
Respects your intelligence and skills.
Check your own automation
Ask yourself these four questions about anything you run.
Your honest answers tell you which kind you have.
In short:
where people stay capable,
automation stays useful.
Your people still know how the work runs.
That keeps the system reliable and the team engaged.
When something unexpected happens, they can deal with it.
Take control of the automation you already have.
Here is how we would do it, together:
-
Build understanding
- Learn what matters to you.
- Map your team and automation.
- Give everyone a shared picture.
-
Add leverage
- Strengthen what works.
- Fix what almost works.
- Remove what holds you back.
-
Gain and keep control
- Teach your team the setup.
- Put the setup to work.
- Keep your people sharp.
Ready to make your automation
make sense?
Esh Sharypin
I am certain the automation that really works makes people feel sharper, rather than replaced.
Throughout my career, I built automation that leveraged people’s potential to outsized results. I also made automation that left people confused, less capable, and worse off than before. Eventually, I learned the difference the hard way.
What I learned is that automation works best when it respects the people using it and makes them better at what they do, and fails when it aims to replace them.
A lot of teams are about to make the same mistakes with AI.
I would rather they did not have to go through it themselves.
Let us find out where this would help you most.
- 13+ years in engineering, product development, and leadership
- Scaled teams and systems 10x at one of Europe's fastest-growing SaaS companies
- Based in Rotterdam and open for business in the EU/EEA
Just curious? The best conversations start that way. Say hi.