.png)
By the numbers:
- Agents in production: ~35 agents in daily production (incl. sub-agents), running daily.
- Enrichment cost: about $0.04/lead
- Contact universe: >1M contacts
1) What’s your role today — and what’s your favorite part of it right now?
I split my day between human touchpoints (early calls, relationship work) and agents (monitoring, fixes, and new builds).
Mornings go to meetings and ‘only-humans-can-do’ conversations; then I check overnight Agent runs, jump on any ‘credit, and error spikes’, and move to new AI automation projects.
I currently have ~35 agents running daily (incl. sub-agents).
My favorite part is when I can build a completely new agent from scratch for a process we haven’t automated yet. The least fun is polishing and fixing old ones.
Balancing competing priorities remains one of my biggest challenges. I often start multiple initial ideas and develop them to ~60–80% completion before pausing and reprioritizing.
2) In one line, how do you define AI Ops from your seat?
Scaling proven GTM tactics with AI Agents — especially the ones traditional automation couldn’t — so they deliver the same results as our best humans, but at 10× scale without the extra headcount.
3) Before → After: the biggest operational change you shipped
The biggest inflection point was inbound lead qualification: we previously had 1 full-time human researching, qualifying, routing, responding to, and booking demos with inbound leads who requested a demo from us.
Now it’s an AI agent doing the work completely with no human interaction — from demo form submission to personalized engagement, negotiating time zones, and meeting booking.
Untangling this process meant embracing agents for the entire end-to-end workflow, then treating reliability like production systems: observe performance, respond quickly to error spikes, and when legacy systems become over-scaffolded, rebuild from scratch rather than patch incrementally.
Case Study #2: Manual and semi-automated CRM enrichment.
Before implementing our CRM enrichment agent: We were doing a lot of CSV shuffling, mass import and export from ZoomInfo/Apollo/Lusha and other lead databases and software, and hand-filling CRM fields in bulk. After: our agent enriches every signup for about $0.04/lead, keeps our >1M contacts fresh with the most up-to-date data, and powers lead scoring and every other sales and marketing downstream automation such as lifecycle marketing emails, outbound sales emails, product emails, and so on.
It’s very reliable with higher quality data than we had before, cheap, reliable, and now the backbone of our GTM.
4) Tell us about an agent you’re proud of (name + job) and how you know it’s “good enough.”
Our outbound AI agents, which write hyper-personalized email sequences based on in-depth research about prospects. They also handle our inboxes completely autonomously, when someone sends us a positive reply.
Here’s how we know they’re good enough: We have received numerous positive feedback complimenting our personalized emails.
But how do we get here? I pilot the emails in stages: I create the system in a way that it’s able to generate a personalized research and then email copy or response to the prospect that only we humans could have done. Then I do this for a few prospects, one by one, introducing different dynamic variables into the system. Then I enroll our first ~25 leads to see if the system works for other dynamic variables. Then I further tweak the system, then I enrol ~100, then thousands. If quality holds at scale, it’s a keep; if not, I fix, improve, and optimize.This staged ramp is what gives me confidence it’s ‘good enough’ in production. After I’m confident in the quality, I optimize for cost at scale.
5) Where must a human stay in the loop — and why?
Relationships. I won’t outsource real conversations to avatars. Agents can prepare me (research, talk tracks), but I build rapport and make the judgment calls. That time brings me joy, and it’s the thing I don’t want AI to replace.
6) A favorite failure → rule story
Probably the biggest mistake you can make with AI agents is not thinking agentic enough and forcing rigid automation rules on agents… you shouldn’t apply the same automation mindset as traditional workflows, because you lose emergent capabilities and risk false positives.
I once forced a pythonic, step-by-step scaffold onto a research agent. It worked, but the system was too rigid for the unstructured nature of our real world, and we lost the emergent capabilities like contextual decision-making — i.e., AI agents using certain AI tools based on the research done so far.My rule since: instruct agents in an objective- and goal-oriented way, not with rigid steps. Then ramp in stages (≈25 → 100 → 1,000 tasks) and observe and optimize in small iterations before scaling. If quality isn’t fine across all tasks, kill the automation and try a different path.
Another favorite: switching to a ‘better’ LLM isn’t plug-and-play, and it can be a costly lesson when you switch an LLM in a production system that handles thousands of tasks and then leave it for a day. Even though costs/tokens improve, the way you instruct different models can shift, and your agents suddenly behave differently.
7) Looking 12 months ahead, what do you hope feels normal — and what will you personally spend more time on?
More manager-style agents coordinating other agents with agents, in a completely autonomous manner will be normal; autonomy grows as LLM economics and intelligence improve.
My focus: designing new growth tactics while agents run the proven plays—pointing toward the “one-person company” north star.
Quick Fire
1. Next thing you’ll automate
- On the product-led-growth side:
— Take Lima — our Lifecylce Marketing Agent — beyond onboarding → always-on personal success coach that drives ‘next best action,’ habit loops, and reactivation nudges for our weekly active users. - Demand capture & gen side:
— Generate more Product Qualified Leads and Accounts, with a hyper-personalised, multi-threaded, account based approach. - Observability & cost reduction:
— Daily/weekly LLM token monitoring dashboards and 10x cheaper-model trials at 80% same quality
2. Tool you can’t live without
CRM Enrichment Agent — keeps >1M contacts fresh for about $0.04/lead, fuels lead scoring, lifecycle programs, and every downstream GTM motion. It’s the quiet flywheel everything else leans on.
3. Any final words of advice?
Don’t sleep on AI. Learn it deeply and learn to co-exist with it — but draw the line: keep your cognition sharp and your relationships human;
‘Don’t outsource your thinking — outsource the execution.’
