Autonomous Software Factories Arrive by 2031 | Luke Alvoeiro
MTS
November 3, 2026
Luke Alvoeiro from Factory breaks down Droid 1.0, introducing a programmatic tool calling engine that generates multi-step scripts to cut context bloat and boost agent reliability.
Luke Alvoeiro
Five years from now, I would imagine humans set out the great picture of what is happening this quarter from an engineering perspective, from a product perspective, and then agents are able to go through and actually execute towards those goals correctly. And that's gonna mean way more multiplayer, way more collaborative experiences so that the human feels just as in the loop as all the other agents in this system, and ultimately also more accountability on software that is actually driving the software factory.
Sophia
Just hit version 1.0, Luke, welcome to MTS.
Luke Alvoeiro
Yeah, thanks for having me.
Sophia
Thank you for joining. You guys are constantly launching new things. I feel like every week there's some new announcement coming out of the factory. Tell us a little bit about Droid 1.0. What is the big announcement with Droid?
Luke Alvoeiro
Yeah, Droid 1 is kind of like our largest agent harness up to date.
And it's like a new revised system that introduces a programmatic tool calling engine into the harness. That's going to drive a lot of major improvements to the accuracy of Droid, the quality of the outcomes that it produces, the speed, the token efficiency, and the reliability of it on any platform.
Sophia
Tell me a little bit about the origin here. So I know one of the cool things about factory, you guys are constantly launching new products, you're identifying what are the specific issues that engineers are facing when they're building out their software. And so what are the big problems here that Droid 1 is set to solve?
Luke Alvoeiro
Yeah, I mean, at its core, so Droid and factory really is able to solve the large use case of how do we get coding efficiency to be efficient within the enterprise setting, where there's repositories of infinite context, where you're coordinating work across maybe thousands of developers, hundreds of thousands of developers in our case.
And you're working with the largest corporations in the world, where you need to give them guarantees around risk, around the efficiency of this tool that they're deploying across the organization and the quality of it overall.
So I think that is the main thing that we're trying to drive with this Droid 1 update. It's making sure that we are able to serve the needs of these customers who have some of the most interesting challenges in the market in terms of how you deploy coding agents effectively. We want to make sure that we can power not only the use case of the human in the loop, but also allow the user to step away from that computer and allow more autonomous workflows. One of our flagship products that we're working on is the Software Factory, right? A trusted, sort of self-improving system that runs 24-7 to build software under control of the user and of this enterprise. And so make sure that the right things, such as the safeguards, the right controls are in place to reach those measurable outcomes. Yeah.
Sophia
Speaking of safeguards, I mean, one of the big areas where a lot of people are considering where the safeguards actually come into place is when it comes to tool calling, especially when you're doing long-horizon tasks or trying to do programmatic tool calling. I know that's one of the big things with Droid 1.0. How are you guys thinking about safeguards around that?
Luke Alvoeiro
Yeah. So at its core, the agent loop is typically involving calling an LLM model, getting a set of tool calls back and then running those according to some engine that you built.
What we really wanted to make sure happened here is as we increase the amount of tool calls that a model can generate ahead of time by using a programmatic engine, which basically creates a JavaScript script that then runs all those tool calls contained within, how do we make sure that that's happening in a safe way such that the user still has the option to manually approve everything that happens at runtime? Because sometimes those tool call parameters, which before were very much known because the model was providing you, now since they're inside of a JavaScript script, they can be more inferred at time of execution. So making sure that that's very safe. Then also making sure that the sandbox environment that this occurs within is very much like locked down, so that Droid cannot have any unintended consequences of operating in this more programmatic way. Ultimately, what that means is making sure that you can give those guardrails in a safe way across all these different models that we provide access to, means that we end up building a harness that it's actually more well suited to the diverse set of tasks and models that then many of our competitors really can create. And that means that you end up generalizing better across the jaggedness of all of these different models and then producing a harness that's way higher in quality and speed and in token efficiency.
Read the other 9 minutes of this episode here.
$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.
Not what you expected? Full refund within 14 days.
Thousands of transcripts fetched by people building searchable podcast archives
Fetch transcripts with code
The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000799606446