from zero to lumenarc (origin story):
Complete meltdown at 22. Walked out of finance. Spent months feeding homeless people and trying to figure out what the hell I was doing with my life.
Eventually stumbled into tech. Taught myself everything. Built ResponseiQ from nothing, raised $2.5M and scaled to hundreds of employees across six countries.
Then covid hit and gutted us (story for another post).
I first got the idea for lumenarc when I realised ai trained everyone to expect instant answers. You ask, you get an answer. Done.
But b2b sales?
Still playing this stupid game of "book a 30-minute call next Thursday." Making prospects jump through hoops just to see your product. Nobody wants to talk to your fking sales rep. They just want to know if your thing works for them.
Lumenarc fixes that.
Everyone's building chatbots that type back at you. We're building ai voice agents that actually sell. And we're testing it on ResponseIQ first. If it doesn't work for our own business, I'm not selling it to anyone else.
Pre-launch now. Building in public (posting about it here). Using it ourselves first.
lumenarc ai
@PawelHuryn Did a bunch of similar tests today and very impressed. Insanely slow when not paying the x10 for rapid, but when your okay to wait superb. I was testing heavily against GLM and it was consistently finding more issues and around 1/4 the cost, but taking 3-4 times as long.
@CasJam@AmpCode love it, my other screen looks very similar right now. Have you played around with plugins yet? I keep trying to find the right usecase.
@sqs 100% keep. Superb defaults, but one thing you can please edit is if I then edit it afterwards to have capitals, rn it edits my edit to lowercase. If I edit to capital manually then that’s clearly for a reason.
10 years ago, cold outbound got me into a boardroom with the CEO of a listed automotive group.
I had zero employees.
One email. One meeting. Me sat across from the CEO and his executive team discussing a rollout across 150 dealerships.
Then they asked the question: "How big is your team?"
Things got awkward fast.
Clearly one guy and his army of outsourced talent wasn't going to cut it for a nationwide rollout. Meeting over.
That's how different cold outbound was in 2015.
Fast forward to now.
I get 50 emails a day. Customised GIFs. AI-generated intros based on my LinkedIn. References to the University of Surrey. Stuff scraped from our website and repackaged as "personalisation."
We all have access to the same tools. Clay. Apollo. Instantly. Pick your poison.
And yes, we use them too. Everyone does. That's the problem.
The conversion rates have plummeted. I'd argue somewhere between 50x and 100x less effective than a decade ago.
Yet somehow there's a 100x increase in consultants and agencies selling these exact services.
The maths doesn't add up.
Here's what's happened in the last two years: AI has turbocharged both sides of the equation.
On one hand, anyone can now send thousands of "personalised" emails in an hour.
On the other, buyers can digest and collate information without ever needing to speak to you. One AI query. A podcast episode. A YouTube comparison video. Done.
These two things collide.
More noise hitting inboxes. More informed buyers who don't need you as much to educate them.
The companies winning now are making it easy for customers to explore on their own terms.
Not forcing them through discovery calls. Not gating everything behind a form. Not pretending a templated email is a relationship.
If you're in software, you probably read Predictable Revenue.
So did everyone else.
That's the problem.
Someone exploited our business and racked up £4,700 in premium hotline charges. In 8 minutes.
Here's the whole story:
Very early days of ResponseIQ.
We had a callback system. Customer requests a call, our system dials them back.
Somebody figured out a flaw and coded an exploit.
Remember those premium rate phone lines?
….The naughty ones. £10 per minute.
Yeah, those still exist. And someone out there still runs them.
They gamed our system to call their line repeatedly during the most obscure hours. Wiped out £4,700 in Twilio credit before Twilio caught it and blocked the numbers.
We got most of it back. But that wasn't the point.
The point was we messed up. We didn't have proper hardcoded guardrails. We assumed people would use the system as intended, but we didn’t quite factor in code-powered robots calling naughty premium lines…
Turns out when you build something on the internet, someone will always try to break it.
GameStop just learned this the hard way too. A YouTuber discovered you could buy a Nintendo Switch 2, immediately trade it back with a cheap pre-owned game, and pocket £50 profit. Repeat infinitely.
Their official response: "We gently remind everyone that our stores are not designed to function as infinite money printers."
It's always a game of cat and mouse.
There's always someone smart out there looking for the vulnerability. Even the biggest companies with the biggest resources make these little mess-ups.
If you've started a software business and don't have a story at least somewhat like this, you're in the outliers.
I'd love to hear yours.
What's the craziest way someone found a hole in your system?
Voice AI adoption isn’t blocked by cost.
It’s blocked by integration.
⠀
According to Deepgram, only 38% of organisations cite cost as a barrier to adopting voice AI.
Meanwhile:
65% cite compatibility with existing AI systems
60% cite integration complexity
⠀
That tells you something important.
Organisations are willing to pay for voice AI.
They’re just not willing to rip out their entire tech stack to get it.
⠀
The problem isn’t the technology.
It’s the ecosystem it has to live in.
⠀
Most enterprises run on a patchwork of systems. CRMs, contact centre platforms, knowledge bases, analytics tools. None of them were designed to work cleanly together.
⠀
Now add voice AI.
It suddenly needs to integrate with all of them.
⠀
Integration isn’t exciting.
It’s often boring to solve.
But in many ways, it’s the only problem that matters.
⠀
Every integration point is a potential failure point.
Every API is a maintenance burden.
Every data transformation is a place where accuracy quietly degrades.
⠀
The vendors who win won’t be the ones with the flashiest demos.
They’ll be the ones who make integration feel invisible.
⠀
If deploying your voice AI requires months of professional services and a dedicated integration team, you’ve already lost the enterprise deal.
⠀
In enterprise software, elegance isn’t about features.
It’s about fitting into what already exists without breaking it.
I get my best work done before anyone can interrupt me.
So I engineered my schedule around it.
Time zone difference. Best decision I ever made.
8-10 AM my time. Nobody else is awake.
No Slack. No emails. No calls.
By the time they're online, I've already done the messy thinking.
I'm in autopilot mode. Don't look my best. But I'm getting stuff done.
I review everything overnight.
News. Data. Customer feedback.
Work through the thinking. Set up clear priorities.
Then gym.
Caffeinated. Focused. Get it done. Come back buzzing. Shower. Eat.
By the time my team logs on, I'm ready to go. They can hit the ground running.
Saturdays are the same. I work Saturday mornings because I'm working ON the business, not IN it.
No interruptions. No reactive work.
Just strategy and experimentation.
Everyone else can have the week. I'll take the quiet hours.
The best work doesn't happen in meetings.
It doesn't happen when you're putting out fires.
It happens when you have uninterrupted time to think.
Two hours of deep work beats eight hours of reactive work. Every time.
Most people don't protect that time. They let their calendar fill up. Then wonder why they're always behind.
I'm a morning person. Some people are night owls. Doesn't matter.
When's your deep work window?