OUR APPROACH
Agentic Engineering
New Playbooks. Same Target.
AI has fundamentally changed how software is written. Here’s how we’re adapting alongside it all.
At Foxbox, agents plan, write, test, and review code on nearly every project. But human engineers stay at the helm. They own the architecture, set the guardrails, and are responsible for what ships.
This page breaks down what it all looks like in practice.
Agentic Engineering, defined.
Agentic engineering is a way of building software where AI agents take on scoped units of work, like planning a feature, writing code, or reviewing a pull request, while engineers direct the work, check it, and take responsibility for the result. Agents work with more independence than an autocomplete tool. Engineers stay accountable for everything they produce.
Six Principles That Guide Agentic Development
Regardless of project size or scope.
The Role of Agents
The Orchestrator: Plans the Work
What it is: An agent that takes a defined objective, breaks it into scoped tasks, and decides what information each downstream agent receives.
What it handles: Task breakdown and sequencing, dependency mapping, context assignment for each builder
Human checkpoint: An engineer reviews and approves the plan before any building starts.
The Builders: Execute in Parallel
What they are: Agents that each own one scoped task, with their own instructions and their own context window.
What they handle: Feature implementation, tests written alongside the code, integration work, migration and replatforming tasks
Human checkpoint: Engineers track progress and step in when work drifts from the plan.
The Reviewer: Verifies Independently
What it is: An agent that checks work it didn't write, against the acceptance criteria, without access to the builder's reasoning.
What it handles: Code review, test and regression verification, security and compliance checks, flagging changes that fall outside the task's scope
Human checkpoint: An engineer signs off before anything merges.
Build What You Couldn't 6 Months Ago
It's a remarkable time to build software. Agents now handle much of the repetitive work that once ate up budgets and timelines, which frees people to focus on what actually drives the success of a product. Tools change, our mindset doesn’t: Aim for indispensable.
Agentic Engineering in Practice: Freshpaint
Freshpaint, a healthcare marketing privacy platform, needed its iOS, Android, and React Native SDKs rebuilt, plus mobile measurement partner (MMP) attribution, without exposing patient data. Our team worked with Claude Code across the workflow: planning before tickets were picked up, building ticket by ticket, having agents self-review before pushing, and running a first-pass pull request review that flagged formatting issues and out-of-scope changes. Engineers handled the final review on every change.
Read the case study