We've shipped 50+ AI products for clients over the past four years. Starting now, we're going to build three of our own — and we're publishing our roadmap because we think the AI development studio model is about to change in ways most studios aren't prepared for.
Why We're Doing This Publicly
Most agencies guard their roadmaps. We get it — you don't want competitors copying your moves. But after four years of building custom AI for everyone from healthcare networks to fintech startups, we've realized something: the studios that will survive the next two years aren't the ones with secret plans. They're the ones with public conviction.
Here's the honest context. Autor is an eight-person AI development studio in Toronto. We charge $150/hr, we only hire senior engineers, and we've never taken outside funding. Our biggest product, Loquent, handles thousands of automated voice calls per month for healthcare and dental clinics across Canada. We built it in eight weeks.
That last part matters. Eight weeks from first commit to production calls. Not because we cut corners — because our team has done this enough times that we know exactly where the time sinks are, and we've built internal tooling to skip them. That internal tooling is what this roadmap is really about.
The Three Things We're Building
1. Loquent Multi-Vertical (Q3 2026 – Q1 2027)
Loquent today handles inbound calls for dental clinics and healthcare offices. It books appointments, answers insurance questions, handles cancellations, and transfers to a human when it hits something it can't resolve. About 82% of calls are fully automated end-to-end.
Here's what we've learned from analyzing tens of thousands of those calls: the hard problems in voice AI aren't industry-specific. The hard problems are latency management, interrupt handling, context switching mid-sentence, and knowing when to shut up and transfer. We solved those problems for healthcare. The actual healthcare knowledge — the appointment types, the insurance logic, the PHIPA compliance — that's configuration, not architecture.
So starting in Q3, we're expanding Loquent to property management, legal intake, and auto service. Three verticals, chosen for specific reasons:
Property management because tenant call patterns are remarkably similar to patient call patterns — high volume, mostly routine, time-sensitive, and currently handled by overworked humans or terrible IVR trees.
Legal intake because law firms already understand hourly billing and the ROI calculation is dead simple: if your intake coordinator costs $55k/year and handles 40 calls a day, and Loquent handles 33 of those, the math does itself.
Auto service because one of our existing clients asked us to, and they're willing to be the beta. Never underestimate "a client asked for it" as a product strategy.
We're not building three separate products. Loquent stays one platform with vertical-specific configuration layers. The voice pipeline, the Twilio integration, the Claude-powered reasoning engine, the fallback logic — all shared. What changes per vertical is the knowledge base, the integration targets (Jane App for healthcare, Clio for legal, Shop-Ware for auto), and the compliance requirements.
Target: three verticals live by February 2027. We'll publish the results — conversion rates, automation percentages, latency numbers — as we go.
2. Autor Deploy Kit — Our Internal Stack, Open Sourced (Q4 2026)
Every AI product we build at Autor starts the same way. We spin up a Next.js frontend, a NestJS API layer, a PostgreSQL database with Prisma, wire up authentication, set up deployment on AWS or Vercel, configure monitoring, and integrate whatever LLM provider the project needs. That process used to take two weeks. We've gotten it down to about four hours.
The tooling that makes that possible — our starter templates, our LLM abstraction layer, our prompt management system, our evaluation harness — has been internal-only for three years. We're open-sourcing it in Q4 2026.
Why now? Two reasons.
First, the AI development market is getting flooded with studios that can't actually ship production systems. They can build demos. They can build prototypes. But they don't know how to handle rate limiting at scale, or how to structure prompts so they don't drift over time, or how to set up evaluation pipelines that catch regressions before users do. Publishing our tooling is a forcing function — it makes us document our opinions about how production AI should work, and it lets potential clients evaluate our thinking before they hire us.
Second, we've watched enough open-source AI tooling projects to know what's missing. LangChain is a framework for building chains. LlamaIndex is a framework for building RAG. But nobody's published a good opinionated starter for "I need to ship an AI-powered SaaS product that handles real users, real errors, and real scale." That's what Autor Deploy Kit is.
It'll include our LLM provider abstraction (swap between OpenAI, Anthropic, and local models with a config change), our prompt versioning system (every prompt is a tracked artifact with evaluation history), our evaluation framework (automatic regression testing against golden datasets), and our deployment templates for AWS and Vercel.
We're not pretending this will be the next big open-source project. But if even 50 teams use it to ship production AI products faster, that's a win for the ecosystem — and frankly, it's good marketing. Every team that uses our deployment kit and likes it is a team that might hire us when they need custom work.
3. Agentic Workflow Platform — Internal First, Then Product (Q1 – Q3 2027)
This is the ambitious one. Over the past year, we've built custom AI agent systems for nine different clients. Document processing pipelines for insurance companies. Research assistants for consulting firms. Automated QA systems for SaaS products. Every one of those projects involved building the same orchestration layer from scratch — the part that manages multiple AI agents working together, handles failures gracefully, maintains state across long-running workflows, and provides observability into what the agents are actually doing.
We're building a general-purpose version of that orchestration layer. Not a framework — a platform. The difference: a framework gives you primitives and says "figure it out." A platform gives you a running system with a UI, logging, deployment, and monitoring built in.
We're building it internal-first. Meaning we'll use it on our own client projects for at least two quarters before we offer it as a standalone product. Every client project in 2027 will run on this platform, which means it'll get battle-tested across different industries, different scale requirements, and different failure modes before anyone else touches it.
The technical bet we're making: the next wave of AI products won't be single-model applications. They'll be systems of specialized agents coordinated by an orchestration layer. The orchestration problem — routing, state management, error recovery, human-in-the-loop handoffs — is where the real engineering challenge is. And it's where most teams are currently duct-taping things together with ad-hoc code.
We don't have a launch date for the external product yet. We'll share progress quarterly.
Why We're Telling You This
Four reasons, and I'll be direct about all of them.
1. Accountability. Publishing a roadmap means we can't quietly pivot when things get hard. If Loquent multi-vertical doesn't ship by February 2027, you'll know. That pressure is useful.
2. Recruiting. We're hiring two senior engineers this year. The kind of engineers we want — the ones who've shipped production AI and are tired of working at companies that demo well but can't operate — those people want to know what they'd be building. Now they know.
3. Client signaling. If you're evaluating AI development studios in Toronto right now, you should know which ones are investing in their own IP versus which ones are just billing hours. We're doing both, and we think that makes us better at client work, not worse. Every product we build internally sharpens the tools we use on client projects.
4. Industry transparency. The AI development studio market is opaque. Clients can't tell who's good until they've already spent $50k finding out. Publishing our roadmap, our technical opinions, and eventually our tooling is our way of making that evaluation easier.
Key Takeaways
Loquent is expanding from healthcare-only to property management, legal intake, and auto service — same voice AI engine, different configuration layers. Three verticals live by February 2027.
Autor Deploy Kit ships in Q4 2026 as open source — our opinionated production AI starter that cuts new project setup from two weeks to four hours.
We're building an agentic workflow platform starting Q1 2027, internal-first for at least two quarters before external launch. The bet: multi-agent orchestration is the next critical infrastructure layer.
We're publishing this roadmap because accountability, recruiting, client signaling, and industry transparency all benefit from openness. Secret roadmaps are a liability when trust is your differentiator.
Everything we build gets published results — conversion rates, automation percentages, real latency numbers. No vague claims.
If you're building something similar, we'd love to hear about it. Reach out at hello@autor.ca or visit autor.ca.
Top comments (0)