Once a system is in production, every day brings questions, support requests, bugs and improvements. Everything comes in through a channel, gets classified, is resolved through the right route and answered with history. Scroll down and walk through the flow, stage by stage.
The flow
Animated diagram for this stage: Four kinds of request (question, support, bug and improvement) arrive through one channel to three actors: ChatGPT analyzes, Claude Code executes and the human team approves data and code.
This is how we work with a client at WITS.
The flow, in short
AI-powered software support and maintenance at WITS is how a system is operated after go-live: every question, support request, bug or improvement comes in through a channel, is classified with AI, is resolved through the right route (system knowledge, data in AWS, or code and sprint) and is answered with history. Six AI agents do the work; the human team approves anything that touches production data or deployments.
Updated · · Reviewed by Samuel Hinojosa, CEO of WITS
E0
At WITS, every request from a client with a system in production comes in through a channel: email (IMAP and SMTP), a Zoho Desk ticket, a Zoho Cliq message or the web system's own chatbot, which answers with OpenAI. Claude Code reads every channel and turns the message into a request with context: who is asking, from which company and what happened.
E1
Before touching anything, WITS classifies every support request as clarification, support, bug or improvement. ChatGPT interprets the message and proposes the category and the route, and a double-lock pre-flight verifies that it is the right company and that its status is active. The classification decides the path: knowledge, data in AWS, or code and sprint.
E2
Questions about a system operated by WITS are answered from its own documentation: manuals, rules and decisions. OpenAI, with RAG over those documents, answers citing the source. The web system's chatbot uses the same base through the API and checks permissions before answering, so each user only receives what they are allowed to see.
E3
When the request is about data, WITS gathers the evidence in AWS: the Aurora database, the S3 bucket, the EC2 logs in CloudWatch and the requests recorded by Nginx. The change is tested first on a local replica and only then applied to the production database, through a controlled channel and with approval from the human team.
E4
If the request is a bug or an improvement, WITS creates an issue in Zoho Sprint linked to the code in GitHub. A development agent writes the change and tests it; the change goes through a sandbox with automated tests and, if everything is green, it goes to production with authorization from the human team. The client receives the answer with the change already live.
E5
The answer to the client goes out through the same channel the request came in: the ticket is answered and closed in Zoho Desk, the team is notified on Zoho Cliq and an email is sent when it applies. Everything is recorded in the client's log, so the next request already has history: what happened, who resolved it and how.
E6
Six autonomous AI agents work on every WITS support request: channel capturer, classifier, knowledge consultant, data operator, development agent and responder. They run end to end. The human team steps in when needed: a change to the production database, a deployment, or a case the agent escalates.
Related
Service
The system is built around the operation; this flow is how we take care of it once it is in production.
See serviceReal case
A multi-tenant SaaS built by WITS: an example of a system in production whose clients ask for support and improvements every day (case study in Spanish).
See caseHow we work
WITS processes from the inside, stage by stage: what the AI does and what the human team decides (in Spanish).
See indexFrequently asked questions
Operations begin: every day brings questions, support requests, bugs and improvements. WITS receives them through a channel (email, Zoho Desk, Zoho Cliq or the system's chatbot), classifies them with AI and resolves them through the right route: system knowledge, data in AWS, or code and sprint. Every answer is recorded in the client's log.
Yes. ChatGPT interprets and classifies every request; OpenAI with RAG answers questions citing the system's documentation; Claude Code executes inside controlled tools, and a development agent resolves bugs and improvements. The human team approves anything that touches the production database or a deployment.
ChatGPT interprets each incoming message and proposes the category (clarification, support, bug or improvement) and the route. First a double-lock pre-flight verifies that the request belongs to the right company and that its status is active, so no data from another client or an inactive account is touched.
Every data change is tested first on a local replica of the database: records are modified or inserted and the result is verified. Only if it is correct is it applied to the production database, through a controlled channel and with human approval. The evidence comes from Aurora, S3, CloudWatch and Nginx.
An issue is created in Zoho Sprint linked to the code in GitHub. A development agent writes the change and tests it in a sandbox with automated tests. If everything passes, the change goes to production with the team's authorization and the client receives the answer with the change already live.
Six autonomous agents: channel capturer, classifier, knowledge consultant, data operator, development agent and responder. They run end to end. A person from the team steps in at three moments: changes to the production database, deployments to production, and cases the agent itself escalates.
In a log per client. Every answer, whether it goes out by email, Zoho Desk or Zoho Cliq, is recorded there. So the next request from the same client already has context: what happened before, who resolved it and how, and nobody starts from scratch.
Yes. The whole flow runs on digital channels: email, Zoho Desk, Zoho Cliq and the system's own chatbot. WITS works remotely, in English and Spanish: questions, support, bug fixing and improvements, with AI agents and human oversight over data and code.
Let's analyze your workflow
We review how you serve your clients today and where the agents would come in. No cost, no commitment.
