Architecture before tools.

We design the reasoning architecture of an organisation — how knowledge moves, how errors get caught, how decisions actually become decisions.

AI is one of the materials we build with. Not the product we sell.

What we sell

Judgment, not calendars.

A consultancy that says yes to everything is selling its calendar. We sell judgment.

Half the engagements we take begin by figuring out what the question is. The work that follows is sized to the answer.

The work

Three shapes the work takes.

The best engineering teams run on a discipline most companies never reached. We bring it to AI — from advisory through to the systems we build and hand back. These three shapes are how it shows up, not a menu to order from.

Reasoning architecture.

We map how the organisation perceives, decides, and remembers — then build the AI systems and the governance scaffolding into that architecture, not next to it. Most programs build the model first and discover the architecture problem at month nine. We start there.

Production AI at the decisions that matter.

The systems hold up when the regulator arrives, when the model deprecates, when the founding team rotates out. SLOs, evaluation harnesses, audit trails, replay. The model is the last piece, not the first.

Forward-deployed engineering.

When implementation is the work, our engineers ship in your repo, your CI/CD, your standups. The team operating the system afterward helped build it. Knowledge transfer is the shape of the engagement, not its final week.

Most AI programs fail at the architecture, not the technology.

Teams overwhelmed by a technology moving faster than their operating model, accepting vendor defaults, spending heavily and auditing none of it. The failure is operational — and it is the gap we close.

95% of enterprise GenAI pilots produce no measurable P&L return
MIT NANDA

21% have redesigned the workflows around GenAI — the biggest predictor of EBIT impact
McKinsey

50% of CEOs report piecemeal, disconnected AI investment
IBM

We ship thirty days from kickoff.

Your legal and procurement set the start date. The build clock — thirty days to a working system — is the part we control.

Week 1

Scope

A long conversation about what success looks like at year two — not just year one. The decision record starts here.

Weeks 2–4

Build

A working system in your codebase, on your infrastructure. Production-grade and owned by you on day one.

Week 6+

Scale

Your senior team understands the system well enough to extend it. Expansion happens with us alongside — then increasingly without us.

Recent writing

From the index.

Procurement The 90 Minutes That Decide Whether Your AI Program Ships. \
A single pre-signing conversation predicts enterprise-AI program survival better than any other procurement signal. Most buyers do not book it.
CAIO The Fractional CAIO Playbook. \
Most full-time Chief AI Officer hires at this stage will fail. The work is real. The shape of the hire is wrong.
SLOs SLOs for Agents. \
Most teams measure model quality and ship hoping reliability follows. It does not. Agents need their own signals.

Leadership

Built inside today's cloud and AI.

Loom is led by Mugdha Pandit and Kevin Dickerson. Between us, we have helped build the cloud, advanced the chipset designs that power modern AI, and shaped how the world's largest companies ship products to billions of people. That is the work we now bring to your AI program.

What we know is which decisions become expensive — and how to architect against them before the first model is trained.

We also support local, mission-aligned small businesses adopting AI, and take on advisory roles and board seats for AI startups and enterprises in transition.

What next

The next conversation is with us.

The first conversation is about what needs to move, who owns the decision, and whether the question is even the one worth answering yet.