# Engineers inside your team until the workflow runs in production. Forward-deployed engineering means engineers join your team and build inside your systems until a workflow runs in production, instead of advising from outside. At GuildBuild, both founders embed for 4 to 8 weeks with a GuildBuild agent team, then hand the workflow to your people or run it for you. ## Two founders and an agent team, working inside your systems. Every embed gets the same pod. We take one embed at a time, so the people you meet on the scoping call are the people who do the work. - **Both founders, as a pair.** They scope the problem with your lead, design the workflow, build it, and review every change before it ships. You talk directly to the people doing the work. - **A GuildBuild agent team.** Agents handle well-bounded engineering tasks under approval gates and cold reviews. Every change they propose is checked before it reaches your systems. - **Your people.** A named lead and the engineers or analysts who will run the workflow afterwards. They work beside the pod, so the knowledge stays with your team. ## Start small. Embed for a few weeks. Keep what works. 1. **Start with an entry offer** (5 days or 2 to 3 weeks). Start with a Microsoft Fabric or Power BI assessment, or an agent team setup sprint. Each ends with a scoped problem, a security review, and an embed plan. 2. **Embed inside your team** (4 to 8 weeks). Both founders work inside your tenant or on your machines with a GuildBuild agent team. The embed ends with one workflow in production and the runbooks to run it. 3. **Hand off or operate** (Your choice at the end). Your people take the workflow over, or GuildBuild runs it for you as a managed service. The goal is not dependency. ## One pod, two kinds of problem. ### Microsoft data For data teams running Microsoft Fabric and Power BI. - Fabric capacity keeps throttling, and nobody can say which workload causes it. - Executives see different numbers in different reports. - A Fabric migration is approved, but nobody has picked the first workload. What an embed ships: One data workflow in production, such as a governed semantic model, a capacity fix with before-and-after evidence, or a first migrated workload running in parallel. ### Agent teams For engineering organizations adopting coding agents. - Developers use coding agents, but there are no shared rules for what an agent may change. - Agent output reaches review with no evidence of what was tested. - Leadership wants a governed rollout, not another pilot that stalls. What an embed ships: One agent workflow your engineers accept, with approval gates, reviews, and a record of what changed and why. ## GuildBuild in numbers Outcomes from projects our principals delivered, results measured in our own agent teams, and how long each step takes. ### Outcomes from delivered projects _Client-reported figures. Clients are not named yet; a reference for each figure is available on request._ - **16.3%** lower average monthly Azure cost year over year, while data and activity grew 30%+. Budget use held at 90.3% on an enterprise Fabric platform. Savings of CAD 80,000+ a year came from Azure SQL storage optimization, orchestration and query tuning. - **2 h → 15 min** average report refresh time, before and after our Fabric and Power BI work. An approximate average across delivered Fabric and Power BI platforms, from incremental loads, model design and query tuning. - **7 weeks** from kickoff to the first governed workload in production. Median across delivered projects, measured to the first scheduled, monitored workload that users rely on. - **4,000+** engineer-hours of delivery work done by governed agent teams (Estimate). 1,020 reviewed pull requests merged in 23 weeks on one client platform, a median of 41 minutes from open to merge. Estimated at 4 hours of human work per change. - **167%** more analysts with direct access to governed data, across 8+ departments. Azure SQL mirrored into Fabric and shared to 8+ department workspaces through OneLake shortcuts: no extra copies or extracts, so analysts outside the central team analyze the same data themselves. - **99.5%** of customer records reconciled to source, from 7 systems into one. Contact centre, billing, Salesforce, field workforce and legacy ERPs joined into one governed customer record, with quality gates before publish. ### What agent orchestration delivers _Measured in GuildBuild’s own agent teams. Each result comes from one published study._ - **28,600** calls run by one 20-session agent fleet. The scale we operate at. 98.1% of the fleet’s input was served from cache. [Source](https://www.guildbuild.ca/insights/multi-agent-fleet-cost-anatomy) - **87%** median context reduction from compaction. Across 196 compactions: far less context for the agent to reread on every later call. [Source](https://www.guildbuild.ca/insights/cache-compaction-economics) - **55%** lower estimated cost for the same perfect score. The right model effort scored 40/40 for less than half the estimated cost of the maximum setting, which scored the same. [Source](https://www.guildbuild.ca/insights/claude-model-effort-benchmark) - **11 min** to hand over the work of six agent seats. After an account switch, the rebuilt team was checked and ready in 18 minutes. [Source](https://www.guildbuild.ca/insights/agent-account-switch-continuity) - **41 of 968** shell calls refused by a guard in one session. Rules enforced in code rather than written in prose, so an agent cannot talk its way past them. [Source](https://www.guildbuild.ca/insights/real-guards-versus-prose) - **3** high-severity bugs caught before a script’s first run. An independent review of an agent-written migration plan; from first review to last fix took 54 minutes. [Source](https://www.guildbuild.ca/insights/review-generated-scripts) ### Delivery you can plan around _Typical lengths. The scoping call confirms them for your situation._ - **5 days** to an agent workflow running on your repository. The Agent Team Setup Sprint ends with one workflow run end to end with your engineers. [Source](https://www.guildbuild.ca/services/agent-team-setup-sprint) - **2 to 3 weeks** to a scored Fabric or Power BI plan. Each assessment ends with a prioritized roadmap or fix plan, a security review, and an embed plan. [Source](https://www.guildbuild.ca/services/fabric-readiness-assessment) - **4 to 8 weeks** to one workflow running in production. The embed ends with the workflow live and the runbooks your team needs to run it. [Source](https://www.guildbuild.ca/forward-deployed-engineering) _Project figures are as reported on those projects and do not promise the same result for your business. Agent-team results each come from a single study; open the linked note for its method and limits._ ## You keep what we build for you. - You own your data, your configuration, and the workflow built for you. - GuildBuild keeps its generic tooling, such as agent team templates and checks, so it can be reused. - The contract names which is which before the embed starts. - At the end, your people take the workflow over, or GuildBuild runs it as a managed service. ## Agents work only where you allow them. Before an embed starts, we agree where your data lives, what the agent team may read or change, and which actions need a person to approve them. The pod works inside your tenant or on machines you control. ## What buyers ask before an embed. ### What is forward-deployed engineering? Forward-deployed engineering means engineers join your team and build inside your systems until a workflow runs in production, instead of advising from outside. At GuildBuild, both founders embed for 4 to 8 weeks with a GuildBuild agent team, then hand the workflow to your people or run it for you. ### How is an embed priced? We quote each engagement after a scoping call, once we know the problem, the systems involved, and how long the work will take. We do not publish rates. Every engagement starts small, with a Microsoft Fabric or Power BI assessment (2 to 3 weeks) or an agent team setup sprint (5 days). ### How long does an embed last? 4 to 8 weeks. The entry offer ends with a scoped problem, a security review, and an embed plan, so the embed starts with an agreed goal. ### Who owns the code and configuration? You own your data, your configuration, and the workflow built for you. GuildBuild keeps its generic tooling, such as agent team templates and checks, so it can be reused. The contract names which is which before the embed starts. ### Do you work on site? GuildBuild is based in Toronto, Ontario, and works with organizations across Canada. Embeds are remote-first, with on-site kickoff and demo days. Delivery arrangements and any data residency requirements are agreed for each engagement. ### What happens after the embed? Your team takes the workflow over with the runbooks we leave behind, or GuildBuild runs it for you as a managed service. ### How many embeds does GuildBuild run at once? One. Both founders work on every embed, so we take one client at a time and agree start dates in advance. --- HTML version: https://www.guildbuild.ca/forward-deployed-engineering Contact: contact@guildbuild.ca · https://www.guildbuild.ca/contact Microsoft status: Microsoft AI Cloud Partner Program member. Not a Microsoft Solutions Partner; not endorsed by Microsoft.