2x Founder & Technical CX leader for companies in complex, regulated industries. I also write about process architecture, sociotechnical systems, and culture.
California, USA
Joined July 2025
- Tweets480
- Following566
- Followers71
- Likes458
Pinned Tweet
I am as delusional as they get. And I am excited about the Informed delusion coming up next with my next project maxwellsdemon.ai 😈
While I have the least to lose, I am going to say something audacious.
My calling is to be Support for humanity.
I have dedicated my life to the craft of supporting people, products, and companies so effectively they need me less over time.
That’s a feature, not a bug.
Dots, the “always on agents”, from OpenAI’s is great for their investors as it means usage is always metered (and invoiced).
The other side of that coin is that users are always paying. And for what? Honest question.
My philosophy on jobs to be done that I share with my teams.
If there is no process, make one.
If there is a process, follow it.
If it is suboptimal, change it.
If you will not change it, follow it.
If you will not follow it, then quit.
Not doing the job is not an option.
Everything is but a system progressing through a lifecycle.
Systems are lucky to terminate at the end of that process, but interference can cause premature termination.
Interfere responsibly.
There are an infinite number of things which can cause a software incident, and each incident can cause an infinite number of consequences.
What is finite is the impact each incident has on a product, which is either working or it’s not.
That’s obviously falsifiable, but what can’t be denied is that during an incident users cannot do the one thing that product is meant to solve for.
Okay, I’m no longer interviewing at any companies.
And so begins the story of my second founding attempt.
Giddy up! Off we go
As AI submits more support requests, and also resolves more of those support requests, volume of inbound emails and calls will scale exponentially.
Systems as they are designed today will fail as they become overwhelmed with information they can’t, or won’t, act on, or discard.
What could your team build if they spent less time rebuilding?
Over my career I've been in more incident retros than I can count. I noticed that asking "5 Whys" often failed in a blameless culture, where the conversation couldn't move beyond the person and on to the organization around them.
After thousands of hours working on, and reflecting on, software incidents at fintech companies, I learned that a productive retrospective answers these 6 questions:
What happened (assurance)
1. What was the trigger?
2. What was the impact?
3. Who was impacted?
What's next (improvement)
4. What is one thing that could have been done to avoid this?
Engineering should implement this.
5.What is one thing that would have allowed us to identify this earlier?
Engineering or Support should implement this.
6.What is one thing that would have improved our response?
Support should implement this.
When done successfully the issue will be:
- Less likely to happen
- Faster to detect
- Easier to respond to
Over time the information learned from each incident, particularly when given to agents, reduces the cost of prevention far below the cost of building fast and breaking things.
But, prevention only works when it is delivered just in time, in such a way that does not require any thought or effort to act on.
Functional examples
- Traffic signs
- Clash detection software
There is still massive opportunity here with AI. I suspect this shift will be difficult though, because business models incentivize increasing dependency more than preventing it.
If knowledge docs in Notion require periodic reviews and verification to remain valid, the laws and legislationin our country should too.
During a review cycle at Modern Treasury, I asked an Engineering Manager and a Manager of Customer Success to be peer reviewers.
Just a few months earlier, in September 2024, I spoke at a company all-hands after I noticed an increasing (and unsustainable) rate of incidents that required escalation and on-call resources across our Engineering and Support teams.
After delivering the remarks below, I spent four months focused on improving escalation processes with Engineering, informed by the expectations from GTM teams and customers, and coached my team to enforce them when automation couldn't.
We set a high bar, and held people to it. Both of which can be unpleasant when done poorly. In the end we pulled it off, and both Engineering and Customer Success teams saw it as a strength.
This is an example of tough love in the workplace, which I've written about on Substack (link in the comments).
---
My script from the September 23, 2024 all-hands
"During sustained periods of incidents, the impact (internally and externally) begins to overshadow the work we do. My role at MT allows me to see how this plays out and I'd like to talk to that briefly.
Externally, customers notice and respect how quickly we resolve incidents, but are becoming fatigued with how frequently they are happening. The REDACTED exercise REDACTED mentioned is an example of how this can escalate. And, this is not an exercise we can get away with doing many times.
Internally, the impact is more discreet. Showing through reduced trust, lower expectations, and doubt. These issues unaddressed can have an insidious impact on culture.
We can rebuild trust internally and externally through small changes and being intentional about what we say, what we build, and what we do as a team. I know this is top of mind for many folks on the EPD end. If you have feedback or ideas about how Support can be a better partner when it comes to incidents please reach out and I’d love to hear from you."
My article on tough love in the workplace is on Substack here: claytonkohler.substack.com/p…
One of the first things I learned after founding a company is that most SaaS products aren’t worth buying, despite the fact so many founders also buy the same products.
Support teams are responsible for triaging and resolving problems which the team did not cause, and which they cannot directly solve. This must be done while also managing the internal and external communication to improve a user's experience while they are having a frustrating experience.
While Al has given Support teams the ability to ship hot fixes, this does not address the root problems, which exist in upstream workflows on Engineering, Product, and Design teams.
With that, the most effective Support teams:
1. Built In-house. Support must be fully bought into individual and group success.
2. Highly technical. Support must have a seat at the table with Engineering and Product.
3. Scale slowly. Support should aim to be the slowest growing team.
4. Expand utility. Support should increase utility through training and tooling.
When done well, the best Support teams make themselves less necessary over time, and so their utility must be continually expanding in order for the team to remain valuable within an organization over time.
I have found that is only possible to achieve when a team is in-house, paid well, and has the technical capability to continually provide value for Engineering and Product teams as products evolve.
Building, and scaling, an effective team begins with creating, and fostering, a culture of psychological safety and a high standard of excellence.
Values of effective support teams:
1. Love and trust
2. Truth and change
3. Humility and conviction
Without these values, and a commitment to service, culture's degrade into a place of modeled apathy, and learned helplessness, from which, in my experience, it cannot recover.
To all of the companies who are dampening the pace of innovation, you better slow the rate of your price increases too, because I am not full paying SaaS fee's for throttled progress.
The lowest level system determines the highest level outcome.
The dangers of AI are real, but the solution is not slowing down. Change, and progress, cannot be stopped and resistence will only increase entropy.
In order for humanity to not only survive, but thrive, with AI we must learn to design and build negentropic systems which can diffuse entropy over time. Anything else leads to heat death.
We must redesign fundamental systems and processes before - not after - offering the world AGI.