Accrual Moves Coding Agents to the Cloud with Runloop

Accounting is a massive industry that has historically bundled labor-intensive preparation with high-stakes professional judgment. Tax preparation alone is a $36.9 billion industry, projected to reach nearly $50 billion by 2030, and is part of a much larger $158.4 billion accounting services market in 2026.

AI is beginning to change that division of labor. By taking on more routine preparation work, AI allows accountants to focus on tasks that require human expertise. H&R Block describes the shift as “technology that elevates expert judgment,” with AI applied to mechanical tasks so “tax professionals can focus more on interpretation, advice, and confidence-building.”

Accrual is building for that future. Its augmented accounting intelligence platform uses AI agents as preparers. These agents handle tasks like organizing client inputs, identifying missing information, generating follow-up questions, and producing draft returns for professional review.  Leading accounting firms, including Armanino, Creative Planning, and H&R Block work with Accrual to transform their work.

Getting coding agents off engineers' laptops

Building that future of accounting puts competing demands on Accrual's engineering team. It means continuously expanding what Accrual's product can do. At the same time, accounting leaves little room to sacrifice quality for speed: deadlines are fixed, accuracy is non-negotiable, and the professionals using Accrual need to trust the output.

Coding agents were already providing Accrual's engineers with greater capacity.

"How do we let engineers parallelize more of their work and get off their laptops?"

was the question the Accrual team asked as they looked to reshape their SDLC. If an agent running on an engineer's laptop could increase individual output, moving that work to the cloud could let engineers delegate multiple streams of work in parallel.

That question led to ARC, or the Accrual Remote Coder: Accrual's internal platform for delegating engineering work to agents that could run remotely and asynchronously. ARC was designed to let engineers hand off multiple streams of work and return to review the results.

But moving agents off engineers' laptops meant explicitly providing things local agents could largely take for granted. A local agent can rely on the engineer's environment, context, and active supervision. Those assumptions break once agents run remotely, in parallel, and across longer-running workflows.

Without a coordination mechanism, complex agent workflows can break down. Agents act on stale data, and long-running workflows lose continuity.

Accrual could have built that agent infrastructure itself, but the team would have to continually keep those systems performant, reliable, and scalable, and taking engineering cycles away from Accrual's actual accounting product. Instead, the team looked for a battery-included platform as the foundation for ARC.

Runloop: the infrastructure for taking agents from local to cloud

“Runloop is the only platform that gives us the full agent lifecycle, the coordination layer underneath it, and the security primitives we need to operate inside regulated environments under a single control plane," — Siddarth Chandrasekaran, co-founder and CTO, Accrual

Runloop Sandboxes replaced the engineer's local machine with isolated cloud environments where each ARC agent could perform its work. Sandbox Blueprints let Accrual define that sandbox once and reproduce it across agents, so new work could start with the tools and configuration it needed already in place.

To solve agent coordination, Accrual used Axons, Runloop's durable event streams. Built into Runloop’s sandboxes, Axons maintains a shared record of what had happened across a workflow. Agents, humans, and external systems could all write to the same history and read it back in order.

‍

‍

Runloop's Broker handled the lifecycle around that state, reliably delivering new work and allowing agents to suspend and resume as events arrived.

Together, Sandboxes, Axons, and Broker is ARC’s foundation, turning coding agents from something an engineer runs on a laptop into persistent workers that could execute remotely, retain history, and continue working asynchronously in the cloud.

ARC becomes part of how Accrual operates

"Being able to throw up sandboxes with Axons meant I didn't have to solve a pile of problems. With any other vendor I would have had to go build that." — Jarrod R., Engineer at Accrual

Using that foundation, ARC agents write code, manage pull requests and stacked changes, respond to automated triggers and webhooks, and continue working asynchronously in the cloud.

Runloop helped Accrual get ARC from conception into production in 7 days, without taking on the infrastructure burden of operating a growing fleet of agents. With execution, persistence, and coordination handled under the hood in ARC, Accrual's engineers could focus on the tools and workflows specific to how they wanted their agents to work.

Today, roughly 80% of Accrual's coding runs through ARC and is the core part of Accrual’s SDLC. Accrual QAs every product flow nightly, determines after each PR whether additional testing is warranted and opens the follow-up work itself, and runs an automated SRE that investigates every error and failed request.

ARC is now expanding beyond engineers. CPAs can propose changes to the tax-preparation logic they work with and route them to engineers for review. Customer-facing teams can have context assembled ahead of meetings, while teams across support, finance, recruiting, and marketing can delegate their own workflows. What started as a way to get coding agents off engineers' laptops has become a way for people across Accrual, technical or non-technical, to build and operate with AI.

Leverage in the age of AI

The principle Accrual brings to accounting also applies to how it is built: focus expertise where it matters most.

For accountants using Accrual, that means spending less time on preparation and more time applying professional judgment. For Accrual's engineering team, it means spending less time building the infrastructure required to run agents and more time building the systems, business logic, and workflows unique to Accrual.

With Runloop providing the execution, persistence, and coordination beneath ARC, Accrual can devote more of its engineering effort to deciding how its agents should work, what they can take on next, to ultimately build a better accounting product.