driftlatched

Cloud architecture and agentic AI for the enterprise.

We connect enterprise systems and business processes to the AI assistants your people already use, then build the secure, governed workflows that put them to work.

Contact

What we do

Architecture and delivery, from the first workflow to the estate it runs on.

  • Agentic AI across the enterprise

    Agents that take on real work across teams and systems: connected to the data they need, limited to the actions they are trusted with, and checked by a person where it counts.

  • Cloud architecture

    Landing zones, networking and identity for cloud and hybrid estates, defined in code so they can be rebuilt, audited and changed with confidence.

  • Governance by design

    Identity, permissions, evaluation and audit built in from the start, so scaling up never means losing sight of what an agent did, or why.

Case study

Tender Watch: an agent for public tenders, reached from Claude

A working system, built and run by us.

Firms that bid on public work check dozens of portals by hand: federal, provincial and municipal, each in its own format, some in French.

We built an agent on Azure that reads open tenders from federal, provincial and municipal sources, around two thousand at a time. It shortlists what fits a firm and proposes pursue, watch or skip, with its reasons. People reach it from Claude through a Model Context Protocol (MCP) connector: what is waiting for review, what closes this week, why it skipped something.

The agent only proposes. A tender becomes a decision when a person records it, and every proposal and decision is kept. The connector sits behind authentication in Azure, and the sources are read, never written to.

A few firms that respond to tenders now use it in their own bidding, in place of checking each portal by hand.

Tender Watch architecture A person asks in Claude. Claude calls an MCP connector over authenticated HTTPS in Microsoft Azure. The connector reaches the tender agent, which reads public tender feeds read-only and proposes pursue, watch or skip. Every proposal and decision is recorded. Only a person records a decision, through the review tool. Claude You ask in plain language MCP over HTTPS, authenticated Microsoft Azure MCP connector status proposals ask review Decisions are made by people only Tender agent An Azure OpenAI model reads each notice and proposes pursue, watch or skip Public tenders Federal, provincial and municipal read-only Record Every proposal and decision is kept
How it fits together. The same shape works for any agent that has to act inside a business: the person asks in Claude, the work runs in your Azure, and nothing consequential happens without a person.

A working agent, making a real decision.

A conversation in Claude. Asked for the latest tenders, Claude reports 33 proposals waiting for review and nine closing within two weeks. For the top one, the agent proposed pursue with a fit of 85, and Claude, checking the source notice, flags that the agent missed it was an Advance Contract Award Notice, so a supplier has already been chosen and the realistic call is to skip.
Taken from a live session. The agent proposed pursuing this one; checking it against the source notice turned up what the agent had missed. A person still makes the call.

Why build it rather than buy a tender tool?

Tender services exist, and each is a portal of its own: another login, another place to check. This puts the work where people already are. The connector speaks the Model Context Protocol, an open standard, so the same capability that answers in Claude today can be offered in Microsoft Teams or any other client that supports it.

It can also run as one step in a larger workflow: an agent that screens overnight, another that prepares a shortlist for the bid team, and people who make the decisions. Each step has only the access it needs.

Tender intelligence runs in Claude today. Teams and the other capabilities use the same pattern, built for your workflow.

Tender Watch is one example. We build the same pattern for other workflows: bring us one you want in production.

How work starts

Usually with one of three things, most often the first.

  • First production workflow

    One workflow, chosen with you, connected to your systems and taken into production. You get it running with review gates, evaluation, monitoring and the documentation to operate it.

  • Architecture review

    A look at an existing estate or AI pilot alongside your team. You get a current-state view, the risks that matter and a prioritised plan for what comes next.

  • Platform foundation

    The landing zones, identity and templates your teams need to run agents at scale. You get it as code your team owns, with the security controls already in place.

Platforms

Microsoft Azure is the platform: the estate, the identity, the data and the services agents connect to. Anthropic’s Claude is the intelligence that works across it. We design the two to fit together.

Microsoft Azure the platform

  • Cloud and AI applications at enterprise scale: landing zones, networking and policy
  • MCP (Model Context Protocol) servers and connectors hosted in your own subscription, behind Entra ID
  • Data and workloads kept in Azure’s Canadian regions, with model hosting chosen to match your residency requirements
  • Hybrid and on-premises with Azure Arc and Azure Local
  • Claude and Azure OpenAI models in Microsoft Foundry, reached by identity rather than keys
  • Infrastructure as code with Bicep and GitHub Actions

the intelligence

  • Claude Enterprise and Team, with single sign-on, admin controls and usage policy
  • Claude Code for engineering teams, with review gates and CI integration
  • Agents that pause for a person before anything consequential
  • Connectors built once and used from Claude, Claude Code and your own agents
  • Pilots measured in production before a wider rollout

Senior-led

Every engagement is led and delivered by a senior technologist, not handed to a junior team.

  • Hands-on

    The architect you speak with is the one who designs and builds the work.

  • Certified

    Microsoft Certified: Azure Solutions Architect Expert. Microsoft Certified: Azure AI Engineer Associate.

The name

On the coast of British Columbia, timber comes out of the forest by water. Logs are gathered into booms and towed down the inlets to the mill. Some slip free. Left alone, a drifting log ends up in open water: lost to its owner, and a hazard to everyone else. So there are people whose work is to go out, find the strays and bring them back to the boom.

Enterprise systems drift in the same way. Configuration drifts from what was approved, models from how they were tested, agents from the scope they were given. We build the latch: the architecture, permissions and checks that keep cloud platforms and AI agents where they belong, and bring them back when they wander.

Contact

Tell us what you are working on and what it needs to do. Most conversations start with one of these:

  • Moving an agentic workflow from a prototype into production
  • Reviewing an Azure or hybrid estate that has drifted from its design
  • Rolling out Claude Enterprise, Claude Code or MCP connectors

We reply within two business days.