Building Behavioural Voice AI. A voice model can speak. An AdaptiveX agent knows how to behave. Building the behavioural layer for enterprise voice agents.
Singapore
Joined April 2025
- Tweets4.2K
- Following292
- Followers343
- Likes2.3K
Great post.
We specialise in outbound lead gen calls.
Both the tech and domain expertise. 20 years in contact centers, 4.5years building conversational voice AI blah blah...
The most important part of the call is the first second. If the customer has to say "hello?" you've lost them.
Our solution has always been to preload the opening line that triggers on first audio from customer.
It creates a seamless opening without skipping a beat.
Vooce AI is now tablestakes. It's the domain expertise and little tweaks that are the difference between a failure or putting $9.9m of pipeline and $3.1m of closed sales into HP's pipeline.
adaptivex.sg/case-studies/en…
ai engineers ship great models and awful first impressions.
cache warming is the easiest, highest-leverage way to cut latency, yet almost everyone ignores it, even in voice ai. that obsession with the first second is one reason why @boardyai stands out!
Brad | AdaptiveX Voice Agents retweeted
just hit $545k MRR... still the only full-time dev
and the app is over 1m lines of code...
in the last 7 days i merged 1,089 PRs
here's the tech stack..
1. over 50 oh-my-pi agents running in parallel on my home linux server
2. 8 Max Claude plans, 1 SuperGrok Heavy plan, and 5 max Codex plans
3. Slack & Telegram to communicate with my bots. I no longer use a terminal or IDE
4. Linear, the central PM tool my bots use for issue tracking. Every engineering request, bug report, alert and error gets posted here
5. Supabase staging database + Railway preview environments for testing
6. Ubicloud runners for GitHub Actions
7. 4 Mac minis as self-hosted runners for iOS builds and 15 end-to-end iPhone journey tests
8. Sentry for error monitoring
9. pstack by @poteto, my main agent uses her orchestrate playbook to manage the other 50 sessions. it never touches code.
10. a custom 5-stage planning harness before any code gets written.
11. customized pr-af and codex for independent code reviews, so that the model that wrote the code never passes it
12. greptile for final code reviews
13. a watcher that tracks every web merge in production for 24 hours, and reverts it on its own if real users start failing
Am I doing something wrong? @ads
I've tried boosting posts a couple of times but it never gets engagement, it seems to not follow the targeted geo and likes are from clear spam accounts.
We created instnct.ai to help businesses focus on what they love while we handle the ringing.
Built on top of our AdaptiveX Enterprise app.
Waitlist open now.
We've been through "it's over" more times than I can count.
It's never as bad as it's seems and there is always an opportunity in these advancements.
Such as this allows you to optimise your workflows to be more efficient in delivering for clients.
man, i'm bummed
honestly most of today i was on fire.
getting a ton of stuff done, all really meaningful.
and then this...and now i'm super depressed.
this is a significant part of my business (our Castos Productions service)...and Codex just does the whole thing.
yeah, this is a very simple example...but it's just like everything else:
This is the worst it's ever going to get.
taking the rest of the day off
We created instnct.ai to help businesses focus on what they love while we handle the ringing.
Built on top of our AdaptiveX Enterprise app.
Waitlist open now.
Tested this today in our production app for internal calls and it's great.
Finally a new turbo model with expressive feature.
Introducing Eleven v4 and Eleven v4 Turbo, our fastest and most emotive voice models yet.
Ranked #1 by Artificial Analysis.
Who doesn't want more speech raining data?! Legends.
Who wants more speech training data?
YODAS v3 is now available @huggingface
It’s 1.1M hours - the biggest audio dataset ever. And the first at this scale with stereo audio at 48kHz, with timestamped transcripts and translations. 100+ langs
hf.co/blog/espnet/yodasv3
CC-BY-3.0
Those of us who use AI in a meaningful way have moved to Opus 5.5.
2 weeks ago it was Astra.
I see a lot of grave dancing on @OpenAI however I keep my Pro plans on both.
The pendulum will swing again and you don't want to be locked out of the Pro plan like they did recently.
25 kills, 13 deaths, one paint-covered rooftop. play RATIO with me api.ratioed.one/s/6vk4b5tjn
Just had Astra email my ex-lawyers asking for the retainer back and if they refuse they'll be hearing from my new lawyers.
Meet my new stylist team... GPT 6 Astra.
While waiting for other agents to finish work they opened my camera, took a pic, created some reference images then had sub agents vote on the cut.
Instructions for barber included.
Back to work.
AI is so good now that if you don't have domain expertise it's almost not worth pursing the idea.
As an example, if we didn't have 20 years contact center experience we wouldn't purse AdaptiveX.
We have best in class tech. So do others.
What they don't have is the domain expertise to create campaigns that scale and produce results.
Most voice agents are tested in polished demos with cooperative humans.
That proves almost nothing.
The AdaptiveX Proving Ground runs the production agent through hundreds of simulated calls with callers who behave like real people.
They interrupt. Mishear. Repeat themselves. Code-switch. Change direction. Refuse the script. Call from noisy environments.
Every call is recorded, transcribed, timed and scored. Exact rules use deterministic checks. Call behaviour is measured statistically. Contextual failures are model-judged and open to human review.
Only agents that clear the release gate go live.
Then the loop continues.
Every production call is analysed within seconds. Drift, failures and recurring patterns feed the next supervised improvement cycle, and the new agent version goes back through the Proving Ground before release.
Your agent should meet its hardest callers before it meets your customers.
adaptivex.sg/platform/simula…
Technical breakdown:
adaptivex.sg/blog/2026-08-20…
I've been quite the past couple of months and for good reason.
We've been hyped focused on deploying our voice agents for HP.
During this time we've managed to add $1.8M (confirmed) into HPs pipeline. This is for one team, there are others.
We did this dialling leads that would otherwise collect dust as it doesn't make commercial sense to deploy a human team.
What sets us apart from other AI voice solutions is our 20 years of call center experience baked into our agents.
We specialise in the hard campaigns. The outbound lead gen from cold leads.
Our agents sounds and act human.
We have a lot of announcements from the past couple of months that I'll share over the next few days.
Head to our website to learn more and have a chat to our agent.
