We Are Orchestrators Now
A plan for the agentic future
Writing code is now a small part of the job. We design systems, direct agents, and ship what they build. The developer who thrives tomorrow is the one who learns to think one level above the code.
Why This Matters
Software development is changing. AI agents are no longer autocomplete tools: they can plan, write, test, refactor, and document code. Agencies that adapt will move faster and free their people for the work that requires human judgment. Those that don't will spend their time doing what machines can do better.
We are choosing to adapt, and to raise our quality bar while we do it.
The Principles We Operate By
Orchestrate, don't just execute. Our engineers define intent and constraints. Agents handle implementation details. The highest-value skill is knowing what to ask for, how to verify it, and when to intervene.
Quality is non-negotiable, and agentic development must exceed it. Tests at every level. Clear documentation. Agentic memory that makes context portable. We do not accept speed at the cost of craft.
Challenge the toolchain, always. No tool is sacred. If GitHub, Linear, or anything else can be replaced or augmented by something that makes us meaningfully faster or smarter, we explore it.
Share what works. Knowledge compounds when it spreads. Wins, failures, and experiments belong to the whole team.
The Plan: Five Steps Toward Full Agentic Operation
Step 1: Equip Everyone
Foundation: give every person the tools to start.
Every team member gets access to first-class agentic tooling: Claude, Codex, and whatever emerges as the standard. No one should be working without AI assistance by default. Onboarding for new tools is treated as essential infrastructure, not optional reading.
Outcome: Every developer is actively using at least one AI coding assistant in their daily workflow.
Step 2: Streamline the Repository Experience
Remove the friction for agents at the code level.
We roll out standardized worktree support across all Rush repositories. Agents need to run in parallel, on isolated branches, without stepping on each other. Nothing else in this plan works at scale until this is in place.
Outcome: Any developer (or agent) can spin up an isolated work environment in seconds.
Step 3: Redirect Shared Learning Toward Agentic Development
We already learn together. Now we point that habit at agents.
Our existing tech lunches shift focus to agentic development: hands-on technique sharing, not general AI news. What prompts work, what orchestration patterns hold up in production, what blew up and why. The sessions stay informal and practical.
Outcome: The team's collective agentic capability compounds week over week.
Step 4: Audit and Evolve the Toolchain
Our tools should amplify agents, not constrain them.
We formally evaluate our current stack: GitHub for version control, Linear for project management, against emerging agentic-native alternatives. This includes trialing tools like VIBE-KANBAN and any other orchestration layer that promises to close the gap between task definition and autonomous execution. We adopt what demonstrably improves throughput and clarity; we don't change for change's sake.
Outcome: A toolchain that an agent can navigate as fluently as a human.
Step 5: Raise the Quality Bar
Agentic speed is only valuable if the output holds.
We write down and enforce higher standards for everything agents touch: comprehensive tests (unit, integration, end-to-end), thorough documentation written alongside the code, and structured agentic memory so that context isn't lost between sessions or agents. Code review now includes reviewing what the agent was asked to do and whether the guardrails held.
Outcome: Agentic-generated code meets or exceeds our previous manual standard.
What Success Looks Like
Three months from now, a new project kicks off. The team defines requirements and architecture. Agents scaffold the codebase, write initial tests, and draft documentation. Engineers review, direct, course-correct, and ship. The quality is higher than before. The timeline is shorter. Nobody spent a week writing boilerplate.
That's where we're going.
Let's build your product
Tell us what you're building. We'll tell you how we'd approach it, what it takes, and how fast we can move.
We'll tell you honestly if we're the right fit. And if we're not, we'll point you to someone who is.