The small teams who treat AI like leverage for their own judgment, not a control system, are going to run circles around the big orgs that still need five approvals to ship a prompt.
When you give a sharp squad full access, clear targets, and permission to move without begging for signoff, all the sludge that slows a typical company disappears and the work finally matches the hype.
I run a small team and there's no legal review. I think once legal is involved, it's back in the cycle mentioned of why small teams are going to run circles around big orgs. Also, for legal, let's build a skill for that. Limit the amount we spend on lawyers. :)
While that’s all well and good that you run a small team, I believe that once you crunched the numbers of how much it’ll cost the company once said pace — a.k.a. a lack of guardrails — causes a serious faulter then there will be a greater understanding and appreciation legal ultimately plays in all of this and/because it’ll be right back to the problem that was just outlined.
Lastly, in the context of emerging agentic AI, what you are ultimately encouraging here is to defer risk in the name of reward. Brave or foolish?…only time can tell.
The agency-over-agents distinction holds. The piece I'd add: even in a flat org, the seam between this team's work and that team's work doesn't go away. The question of who holds it does. The role that carries intent across seams without becoming the gatekeeper is the function I keep finding missing in AI rollouts. I've been calling it stewardship. Not the middle management you're rightly critiquing. Adjacent to it.
You’re pointing at the #1 issue facing enterprise businesses. Their size, org structure and processes block innovation.
I’d not start with R&D, usually they are the ones eager to try AI and this provide little value (unless your a software business).
Having start SMEs in each department and across functions play with AI would not hurt. They’d get a first hand experience of what it can do and limitations.
Agents are probably a phase 3 or 4. You can’t automate and delegate what is not standardized and documented. If it was, it’d already be (semi-)automated. You didn’t need agents for that.
imo each company needs to build a common AI trust language (my version are 'ai trust heuristics' https://auxfirst.com/heuristics.html) but right now I see more of a 'everything everywhere and all at one' approach which is (unfortunately) jeopardising trut in AI
The trust framing lands. In smaller services businesses I see the same root cause in a different coat: founders won't let AI near anything client-facing because they don't trust the output, so it gets relegated to toy tasks and never compounds. The unlock isn't a better tool — it's narrowing the scope until the output is verifiable, then widening it once trust is earned. Same way you'd onboard a person.
But I have to say, for me and probably a lot of founders and builders in this space, making fast and cheap decisions isn't half as important as making good decision. And good decisions take time no matter how many agents you have working for you.
I know Loveable (and Anthropic and OpenAI and Google) has an incentive to say this, but the obsession with "going faster" can only increase the amount of slop out there – when what we all want is more thoughtful, more intentional and more delightful products to use.
Not to sounded blithely cynical but y’know…economy of words: Such companies will simply replace these same //needy// employees with engineers and not these same trained, non-technical employees.
Found the part about having a parent for each agent to give it the right context really fascinating! One thing I'm curious about - In larger organisations, there may be resistance from people around sharing their tacit knowledge for fear of being replaced, how do you encourage people to share their full context then?
thanks for the article! you totally can hand someone the best model on earth and they'll still sit waiting on a slack approval. the permission structure often takes a bunch of energy and time
Great read and good insights. What I continue to see as companies leverage more AI to go faster and are able to manage the gates effectively, is a quick acceleration of great things. However, they next have to figure out how to manage wasted spend, often from redundant AI efforts or lack of context causing AI churn. Ultimately, many run into a lack of creativity problem. If your company lacks vision and your teams don't know what to build faster with AI, you're just spending credits and all too often eliminating roles under the unfair label of "AI caused this". The teams who get behind a vision, understand what's possible, and are unleashed to go after it, quickly have a plan and execute faster.
The small teams who treat AI like leverage for their own judgment, not a control system, are going to run circles around the big orgs that still need five approvals to ship a prompt.
When you give a sharp squad full access, clear targets, and permission to move without begging for signoff, all the sludge that slows a typical company disappears and the work finally matches the hype.
Yes but what happens when the lack of review here finally catches up? Legal would have a meltdown.
I run a small team and there's no legal review. I think once legal is involved, it's back in the cycle mentioned of why small teams are going to run circles around big orgs. Also, for legal, let's build a skill for that. Limit the amount we spend on lawyers. :)
While that’s all well and good that you run a small team, I believe that once you crunched the numbers of how much it’ll cost the company once said pace — a.k.a. a lack of guardrails — causes a serious faulter then there will be a greater understanding and appreciation legal ultimately plays in all of this and/because it’ll be right back to the problem that was just outlined.
Lastly, in the context of emerging agentic AI, what you are ultimately encouraging here is to defer risk in the name of reward. Brave or foolish?…only time can tell.
The agency-over-agents distinction holds. The piece I'd add: even in a flat org, the seam between this team's work and that team's work doesn't go away. The question of who holds it does. The role that carries intent across seams without becoming the gatekeeper is the function I keep finding missing in AI rollouts. I've been calling it stewardship. Not the middle management you're rightly critiquing. Adjacent to it.
You’re pointing at the #1 issue facing enterprise businesses. Their size, org structure and processes block innovation.
I’d not start with R&D, usually they are the ones eager to try AI and this provide little value (unless your a software business).
Having start SMEs in each department and across functions play with AI would not hurt. They’d get a first hand experience of what it can do and limitations.
Agents are probably a phase 3 or 4. You can’t automate and delegate what is not standardized and documented. If it was, it’d already be (semi-)automated. You didn’t need agents for that.
I think the Reed Hastings approach to company leadership is just hands down the best approach. Hire exceptional talent and trust them.
imo each company needs to build a common AI trust language (my version are 'ai trust heuristics' https://auxfirst.com/heuristics.html) but right now I see more of a 'everything everywhere and all at one' approach which is (unfortunately) jeopardising trut in AI
The trust framing lands. In smaller services businesses I see the same root cause in a different coat: founders won't let AI near anything client-facing because they don't trust the output, so it gets relegated to toy tasks and never compounds. The unlock isn't a better tool — it's narrowing the scope until the output is verifiable, then widening it once trust is earned. Same way you'd onboard a person.
Always enjoy your writing.
But I have to say, for me and probably a lot of founders and builders in this space, making fast and cheap decisions isn't half as important as making good decision. And good decisions take time no matter how many agents you have working for you.
I know Loveable (and Anthropic and OpenAI and Google) has an incentive to say this, but the obsession with "going faster" can only increase the amount of slop out there – when what we all want is more thoughtful, more intentional and more delightful products to use.
Not to sounded blithely cynical but y’know…economy of words: Such companies will simply replace these same //needy// employees with engineers and not these same trained, non-technical employees.
((What do?))
Found the part about having a parent for each agent to give it the right context really fascinating! One thing I'm curious about - In larger organisations, there may be resistance from people around sharing their tacit knowledge for fear of being replaced, how do you encourage people to share their full context then?
thanks for the article! you totally can hand someone the best model on earth and they'll still sit waiting on a slack approval. the permission structure often takes a bunch of energy and time
Hello how are you doing today
https://tomzrubecky.substack.com/p/the-vision-for-new-hr?r=10r8xy
Great read and good insights. What I continue to see as companies leverage more AI to go faster and are able to manage the gates effectively, is a quick acceleration of great things. However, they next have to figure out how to manage wasted spend, often from redundant AI efforts or lack of context causing AI churn. Ultimately, many run into a lack of creativity problem. If your company lacks vision and your teams don't know what to build faster with AI, you're just spending credits and all too often eliminating roles under the unfair label of "AI caused this". The teams who get behind a vision, understand what's possible, and are unleashed to go after it, quickly have a plan and execute faster.
Usually an operating model problem