AI agent development services
What we build
Customer-facing agents
Support agents that read from your knowledge base and ticketing system. Sales agents that qualify leads against your CRM rules and book meetings. Voice and chat, with deterministic fallbacks when the model is uncertain.
Internal operations agents
Project status, document drafting, meeting summaries, code review, invoice matching, supplier onboarding. We run this pattern ourselves with Owtoo, our internal AI assistant.
Content and data agents
Product enrichment, catalog cleanup, classification, translation, multi-channel content generation. Our WriteText.ai product is one of these, built for ecommerce.
Workflow and integration agents
Agents sitting between your ERP, CRM, PIM, and ecommerce platform. Reconciling records, triggering actions on threshold events, escalating exceptions to the right person.
Autonomous research and analysis agents
Open-ended work with clear inputs and outputs but unpredictable steps in between. Competitive research, document assembly, multi-source comparisons.
How an agent project runs
Discovery. We define the agent’s job in a single sentence, list the systems it has to read from and write to, and agree what success looks like.
Proof of concept. We build the agent against your real data with an evaluation set you can read. You see retrieval accuracy, action accuracy, and where it gets things wrong. By the end of this phase you have a working demo and a clear yes-or-no decision.
Production build. Integration with your systems, authentication, audit logging, error handling, rate limits, fallback paths, and the evaluation harness that catches regressions when models or data change.
Operation. Monitoring, prompt tuning, model upgrades, gradual handover. Or we keep operating it, with reviews against the original success metrics. Your call.
What you get
Code and prompts you own
When we ship, the agent's code, prompts, integrations, and deployment are yours. No proprietary platform you can't leave.
Built on your stack
We integrate with the systems you already run. Your CRM, ERP, ecommerce platform, authentication, data warehouse.
Monitoring from day one
Every agent ships with logs, evaluation tests, and the alerts that tell you when the model starts misbehaving. The system doesn't degrade quietly.
Model-agnostic engineering
We pick the model that fits the project and design the architecture to swap it when a better one ships. You're not locked to a provider.
When an agent isn't the answer
If the task is deterministic. A workflow with five clear if-then branches needs a script, not an agent. The model adds latency and unpredictability for no gain.
If the cost of a wrong answer is very high. Financial transactions, legal commitments, medical decisions. We can build agents in adjacent workflows, but the irreversible step belongs to a human.
If the data isn’t there yet. RAG and agents both run on what they can read. If your knowledge base is empty, contradictory, or out of date, fix that first.
Commonly asked questions
Which AI model do you use?
Whichever one fits the project. We test on your data before committing, and we design so the model can be swapped when something better ships. The field moves too fast to lock in.
What happens to our data?
For most projects we use commercial AI providers with data-retention-off settings and no training on customer data. For projects with strict residency or confidentiality requirements, we set the deployment to match.
Can you take over an agent another team built?
Yes, with an audit first. We review the prompts, the integrations, the evaluation set if there is one, and the operational telemetry. Sometimes the right move is a rebuild. Sometimes it’s a few weeks of cleanup. We tell you which.
Can the agent be voice-based?
Yes. Voice agents have a few extra moving parts (telephony, speech-to-text, text-to-speech) but the underlying agent work is the same.
Do you build AI features into projects that aren't AI projects?
Yes, often. Our regular software development work increasingly includes AI features where they fit. If you’re not sure whether your project counts as an “AI project,” it doesn’t matter. Same team, same approach.
Book a consultation to explore how AI and 1902 Software can transform your next ecommerce project — faster, smarter, and more cost-efficient.