“Has anyone approved this yet?” Process automation where AI agents do the steps and your team decides.

For work that crosses departments and tools: a purchase request, a client onboarding, a weekly report. The agent reads the request and the documents, checks them against your records, prepares the case, and stops where a team member has to decide.

Free, with no commitment. A senior engineer replies within one working day.

For
Processes that cross departments and wait on approvals
You get
The process running end to end, with approval steps and a record of every decision
Approvals in
Microsoft Teams or Slack
Price
One fixed quote for the project
Your systems
Stay as they are. Nothing is replaced or migrated

Process automation with AI agents means handing over a whole recurring process, not a single question. The agent reads the request and its documents, checks them against the company's systems, does the routine steps, and brings each decision to the right team member in Teams or Slack with the facts attached. Deep Blue Coding builds these processes, on Claude by default, with an approval step wherever a team member has to decide, and a record of who decided what.

Where a process gets stuck today

A request arrives in one line. Someone opens three systems to find out what it really means, reads a contract to check a price, and writes an email asking for approval. The approver gets a message with no context, asks two questions, and answers the day after.

None of the steps is hard. They are slow because someone on the team has to carry the case from one tool to the next.

What you get

The process, mapped and agreed
Every step, who decides what, and what counts as an exception, written down with the people who run it.
The agent doing the steps
It reads requests, PDFs and scans, looks things up in your systems, notices what does not match and prepares the case.
Approval messages
The approver gets a message in Teams or Slack with the facts, the recommendation and two buttons.
A record of every decision
Who approved or rejected, when, and what they were shown.
Exceptions sent to your team
When something is missing or unclear, it goes to a team member with the reason.
Review mode
Before going live, everything is prepared for review and nothing is sent on its own.

An example: a laptop request that does not match the contract

The process

A manager asks for laptops for new hires. Operations checks stock, finance checks the supplier contract, and someone senior approves.

Where it goes wrong

The request is vague, two laptops are already in stock, and the supplier's quote is above the contract price. Nobody notices until the invoice arrives.

What the agent does

It reads the hiring plan to work out how many are needed, finds the two in inventory, sees the quote is 14% over the contract, and sends the approver the corrected order to approve.

How approvals reach the right person

An AI assistant does not come with a way to send a team member an approval with buttons and carry on once they answer. That part is what we build, and it has three pieces:

A connector
With two actions: request an approval, and check its status.
A messaging app in your tenant
A Teams bot with cards and buttons, or an internal Slack app. If you already use Power Automate, its approvals can do this job.
A small server
It checks who pressed the button, keeps the record and resumes the process. It can run on your infrastructure or be hosted by us.

How it runs

  1. Pick one process

    Usually the recommended one from the assessment.

  2. Map it

    With the people who run it, including the exceptions they handle from memory.

  3. Build it on past cases

    We run real cases from the last months through it and compare with what your team did.

  4. Review mode

    It runs alongside your team. Everything is ready for review, nothing goes out on its own.

  5. Go live with approvals

    The approval steps are switched on in Teams or Slack.

  6. Add the next process

    The connectors and the approval pieces are already there, so the second one costs less.

What we need from you

  • The people who run the process today, for the mapping and the review.
  • Past cases to test with.
  • An administrator for Teams or Slack to install the approval app.

What it costs

One fixed price for the project, quoted after a free assessment. The first process carries the approval pieces; later ones reuse them. A process that continues after an approval calls the model through its API, which the model's vendor bills by usage. We estimate that cost in the quote. See engagement models.

Questions buyers ask

Can you add AI to a product or system we already have?

Yes, and it is part of the same service. We build AI into software you already run: search over your documents, document understanding, classifiers, computer vision, predictive models and assistants inside your own application. The engineers who build it are the ones who maintain it.

Does it have to run through a chat?

No. A chat suits requests written in plain words, and it is one way in. We also build custom software and custom workflows: a web application with its own screens, a portal for clients or suppliers, or a flow that starts on its own when an email, a form or a new record arrives. The connectors and the approval steps are the same.

Which processes are good candidates?

Recurring ones where a team member reads documents, looks things up in several systems and then asks someone to approve: purchase requests, client onboarding and KYC, weekly and monthly reports, supplier checks, contract reviews.

Can the agent act without anyone approving?

Only where you decide it can. You choose, step by step, what runs on its own and what waits. Anything hard to undo, such as a payment or a message to a client, waits for a team member.

What happens when the agent is not sure?

It stops and sends the case to a team member with the reason: a document is missing, two records disagree, or the request does not fit any rule it was given.

Does it work in Microsoft Teams and Slack?

Yes. The approver gets a message in the tool they already use, with the facts and two buttons. In Slack it is an internal app in your workspace. In Teams it is a bot installed in your own tenant, which we test against your tenant before we commit to dates.

What does the record of decisions contain?

For each approval: who was asked, what they were shown, what they answered and when. It is stored by the approval service, which can run on your infrastructure or be hosted by us, and it can be exported.

Last updated 18 November 2026.

Start with one process.

The report your team dreads building, or the approval that sits in an inbox for days. Tell us about it and a senior engineer tells you whether it can be automated and what it would take.

Email social@deepbluecoding.com
We reply within one working day.