Super Agent With a Forward-Deployed Human
Give the whole company one shared agent, owned by a human who keeps it working — not personal agents for everyone.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 3
- Confidence
- 90%
An organizational model for deploying AI agents at a company. Instead of every person maintaining their own agent (which breaks constantly and demands too much fiddling), you run one shared 'super agent' for the whole company, owned and gardened by a forward-deployed engineer whose job is to keep it working for everyone. As models get more independent, you push agents down into specialized team agents. The load-bearing principle: automation is a lie — every agent needs a human who cares about it.
Origin
Dan Shipper's revised view after Every adopted personal agents (via Open Claw) company-wide and found them too high-maintenance; he cites Shopify and Ramp as companies running a single company-wide super agent.
Core principles
- 01Automation is a lie: every agent needs a human on top making sure it works
- 02An agent is only useful while a human personally cares about it; sever that connection and it stops being useful
- 03Start centralized (one agent at the top) and specialize downward as models get more independent
- 04The forward-deployed engineer is a real, durable job created — not eliminated — by automation
How to run it
- 1
Start with one agent at the top
Stand up a single company-wide agent (often reachable in Slack) that many people use, rather than provisioning a separate personal agent per employee.
Pro tip Anchor it to a job everyone in the company needs, like data requests, so adoption is immediate and the value is obvious.
Watch out Personal-agent-for-everyone sounds appealing but collapses under maintenance load — most people won't or can't SSH in and babysit a fragile harness.
- 2
Assign a forward-deployed human owner
Put a forward-deployed engineer (or someone with that profile) in charge of keeping the agent working, adding context, catching bad output, and fixing breakage for the whole company.
Pro tip Much of this person's work is conversational — talking to the agent in Slack, questioning its dumb moves, and correcting it — not just writing code.
Watch out The minute no human cares about the agent, it stops being useful; don't deploy an agent you won't staff an owner for.
- 3
Specialize downward as models improve
Once the top-level agent works, spin out specialized team agents; as models get better at running independently, push toward more personal agents that need less internal fiddling.
Pro tip Build a system that lets less-knowledgeable people use the agent safely, rather than just reacting when it breaks.
Watch out Don't assume the models are autonomous enough yet to skip the human layer — for now, more automation has meant more human oversight work, not less.
In the wild
Every runs an internal agent called Claudie that operates its whole consulting practice. Nitesh, a forward-deployed AI engineer, spends most of his time in Slack talking to it — correcting it when it does something dumb — rather than only writing code.
→ The agent runs the practice while a single human keeps it on track, illustrating a durable new role created by automation.
A large model company built one data-science bot, hooked to the data warehouse with permission-awareness, that any employee can query. A dedicated team keeps it accurate so the data-science team no longer fields every basic request.
→ Data scientists are freed from trivial queries to work on deeper questions; without the team maintaining the bot, they'd 'hate their lives.'
Common mistakes
Giving everyone their own personal agent too early
Personal agents break constantly and demand SSH-level fiddling most people won't do; once someone stops maintaining theirs, it becomes useless, so the model doesn't scale until models are far more independent.
Believing automation removes the human
Every automation needs a human on top making sure it works well; treating an agent as fully autonomous leads to unmonitored, degrading output and unfixed breakage.
Is it for you?
Best for
Founders and ops leaders rolling out AI agents across a company who want reliable adoption without per-person maintenance chaos
Not ideal for
Solo operators or tiny teams where a single person can own their own agent, making a central super-agent structure unnecessary
From the transcript
“Automation is a lie”
“in order for an AI agent to be useful right now, it really needs a human who cares about it”
“you basically set up a forward deployed engineer or someone with that sort of profile who's responsible for making sure that that agent is working…”
“the mechanism is agents need people who care about them”
From the episode
The AI paradox: More automation, more humans, more work
Dan Shipper