We Need To Talk
A variety of updates across our ecosystem: but first, an actual warning if you aren't taking AI seriously [at your call center]
Back to normal call center stuff. We're talking project updates mostly, but before we jump into those: we need to talk.
Yes, one of those kind of talks.
Here's what you can expect if you continue on:
- We need to talk
- Starbase updates
- Mission Control updates
- Call Theory Updates
- IS Scripting Updates
- Call Center Village/Conference Updates
- Scripting Sessions: Feedback Wanted
We need to talk
As I brought up previously, the "moat" that we've enjoyed as an industry in voice latency isn't quite a moat anymore.
In the last several months, SOTA (State Of The Art) AI models/systems have enabled low-latency communications over telephone systems (to a degree acceptable by even me, a skeptic), allowing an explosion of "AI Call Centers" and other replacements for our industry at massive scale and a much lower price point than live agents.
By latency, I'm referring to the round-trip time it takes to process human voice input and respond with AI generated output.
Example AI Agents
Genesys Cloud Virtual Agent
Amtelco Ellie Virtual Agent
ElevenLabs Conversational AI
Retell AI Voice Agent
Twilio AI Assistants
Relay Hawk AI Agents
SignalWire AI Agent API
Bland.ai Conversational AI
Voiceflow AI Agents
These aren't referrals or paid endorsements. Just concrete examples for those reading. There are many, many, many more options out there.
What this means (I think) is that you need to start sounding the internal alarm if your call center isn't at least addressing the "AI Boom" (or "AI Bubble", depending on your outlook.)
Be Prepared. – Boy Scouts of America
AI Development Pillars
While I'm still developing my own AI morals and how they apply to application development that I'm responsible for, I wanted to share a few items that I'm personally and professionally trying to follow:
- Focus on improving the worker experience, above all else.
- Avoid reliance on services and build local-first.
- The context isn't safe.
Focus on the worker experience
By the worker experience, I mean elevating and improving a live operator's ability to handle calls without simply replacing them with cheap robotic labor. More broadly, use AI to figure out real problems for your employees so you can move towards the 4-day work-week, right? Right?
We aren't striving to actually become our satirical company AnswerTarget.
Avoid Reliance On 3rd Party Services
I'm possibly distorting the concept of local-first for my own benefit, but I mean the ability to use the software you download without needing a 3rd-party company to provide a service to you. Use services for backups, testing, and development – but take the time to build a sustainable technology stack for risk-mitigation against price increases of services.
Relying on a technology that can't be ran locally by most people? That's a service.
If your entire business model exists on top of a 3rd-party service, it's subject to the pricing and direction of that service. That's not a position most businesses want to be in.
The Context Isn't Safe
The final point is primarily an ongoing reminder that (at this point in the AI ecosystem) anything you put in the context of an AI Agent can likely be exploited through prompt engineering.
So build accordingly.
Starbase Updates
I talked about Starbase "crashing and burning" when I tried to make it compatible with IS 5.7. I considered for a moment giving up, as maintaining multiple versions of this product for different Amtelco server versions isn't something I had calculated into the equation originally.
Probably something dumb to overlook, but without any vendor support from Amtelco I'm limited to what I can test against until a customer is able to access it.
However, after some discussion with customers and digging a little deeper into the 5.7 changes that caused problems, we've determined that we can automatically identify which version of IS you are using and then load the proper schema operations on the fly.
A screenshot or two
We will still have to test new versions explicitly as they undoubtedly will continue to add new billing modifications into future, but now we don't have to worry about explicit version differences. This took a lot of effort but I think it will likely save a lot of version update issues moving forward.
Publish coming as soon as we get all of our tests passing.
Mission Control Updates
We've (finally) got working WCTP support integrated into Mission Control, allowing anyone using our legacy WCTP gateway to migrate and decommission the older software.
With this update, we also updated our notifius/php-wctp library to be compatible with php8.4
Next, we've published a demo MCP over SSE protected by API token authentication. It's like a hello_world version of of MCP so I can do some real-world testing before we start building out additional MCP tools.
This feature requires a modification to the underling nginx proxy configuration to enable SSE.
The first (real) MCP tool will be a vCon representation of Genesis calls.
Bugs & Fixes
It's not all sunshine and rainbows. We're working on resolving a few bugs:
- A SAML bug we introduced in the last version,
- Faxing support for Infinity breaks UC Supervisor redirect/resend
- A broken Summary-to-Image pipeline for People Praise submissions
We're testing updates on these as we speak!
Call Theory Updates
Call Theory, our customer portal, has a few updates as well.
We've added a new sub-site for AI Agents that we will build out as we learn. With this change, we've also re-arranged the navigation section at the top so that the Amtelco documentation and reporting sections are grouped together.
Did I mention we moved all of our Amtelco documentation into it's own section? This means that some of your bookmarks may need updated.
You'll also see a new drop-down to view the Starbase and Mission Control documentation (these are just being consolidated into a drop-down navigation.)
Finally, we've updated our management back-end to better accommodate those who wish to pay by check instead of credit-card or ACH. Thanks for your patience on this one.
Expect these changes to go live over the Labor Day weekend.
IS Scripting Updates
I have one tip you probably already know, and a fun update on automated Intelligent Series Scripting.
Copy & Paste w/ Advanced Expressions
I learned from a customer that you can Copy & Paste working Advanced Expressions between elements with a simple process:
- Click the "Copy" button in the first Advanced Expression editor
- Close out of the Advanced Expression editor
- Open the second Advanced Expression editor
- Click the "Paste" button the the second Advanced Expression editor
Mind. Blown. (Thanks Jenny!)
A DSL for Intelligent Series Scripting
We've been working on using AI tools to try and help speed up the Amtelco Intelligent Series scripting process. It's something that you heard us announce at NAEO conference earlier this year. Well, we've made some progress!
We've gotten familiar with setting up training data, and we've been able to use the plethora of existing Intelligent Series scripting that we've created over the years to really get this moving.
As part of this effort, we've designed a Powershell-based DSL (Domain Specific Language) that allows us to pragmatically generate many elements into an .iif file compatible with import.
It can also take an existing script, add/edit elements, and re-compress everything back into the .iif file.
Yep, you read that right. We can pragmatically generate common scripting to quickly scaffold scripts, and use the same DSL to edit existing .iif files.The limitations currently are mostly around complexity (nesting) but those details will get sorted out with time.
But we've also designed a tool that will convert your Intelligent Series scripting into a variety of different forms that can be used stand-alone as HTML pages or as import formats for converting to other systems pragmatically. (Or just as backups when using Emergency Agent.)
My immediate use-case was for a specialized JSON format I can use for setting up training data for NLP.
The only item left to make this a Natural Language to Intelligent Series script tool is to finish generating training data and training (fine-tuning) our model on that data.
The current plan for this tool is for us to keep it internally until we've got it humming along nicely, at which point we'll use it to augment our own scripting services.
I haven't decided if this will eventually become a product or an open-source tool. I had hoped the community would help in this effort by submitting scripting examples for training, but nobody was interested.
So now, I'm having a Little Red Hen moment, honestly. I'll let you know where I land as this tool matures.
Conference Updates
We're still networking and building out our Call Center Village schedule by volunteering an attending a few regional events:
- We'll be attendees at BSides Columbus
I think our talk got rejected, it's still "In Evaluation" but we're only 3-weeks out.
- We are officially volunteering at GrrCON
- We'll be attendees at BSides Cleveland
- We are attending QueenCityCon (but also volunteer and village submitted)
We're also hoping to get into BSides Milwaukee in Spring 2026.
Scripting Sessions: Feedback Wanted
A simple question for those who attend our weekly training on Intelligent Series scripting:
Do you want these sessions recorded and put on the website like the original Space Camp sessions?
We can avoid recording any troubleshooting we do ahead of time.
See you next time.