Operational AI
Your team is doing work the software should be doing.
Built by Berry is the operational AI firm — we ship the systems, the agents, and the training to run them. First working system in 30 to 60 days.
Best fit when ops is rebuilding records by hand, leadership cannot trust the numbers, or the AI you bought never made it past the early adopters. Industry does not matter. Operational complexity does.
What we ship
Swarm — our open-source agent orchestration framework for Laravel. The same package we use on every engagement.
See SwarmOne real workflow
Account onboarding
Before — Sales closes a deal. Ops spends ~40 minutes per account rebuilding the record across CRM, billing, and onboarding tools.
After — One workflow handles the routine path end-to-end. Agents draft, route, and reconcile. Humans approve exceptions only.
Time per account
40 min → 4 min
Handoff errors
Routed, not lost
Data drift
Reconciled by default
What we mean by operational AI
AI that does the job. Where the job lives.
Operational AI is AI that actually shows up at work — embedded in the systems your team runs the operation on. Not floating in a chat window. Not stranded inside someone else's product.
It does the multi-step jobs your team handles by hand today. The kind that earns its keep in operating metrics, not pilot reports.
Example — workflow execution
Triage a support inbox the way your best agent would.
Read the ticket. Pull context from CRM and billing. Draft a response, attach the right account notes, and either send it or route it to the human who should see it. Logs every step. Learns the patterns your team already runs.
Example — embedded judgment
Flag the deal that needs ops review before it closes.
Watches every new opportunity for the patterns that historically caused onboarding chaos — odd contract terms, non-standard regions, missing data. Pings the right person before it becomes a Monday-morning fire.
Featured engagement
From spreadsheet rebuilds to one operating workflow.
A ~50-person B2B services firm. Anonymized; the work is sensitive.
The situation
Every new account triggered three hours of cross-tool data entry across CRM, billing, and the onboarding tracker. Mistakes surfaced two weeks later as billing disputes and missed SLAs.
What we built
A workflow system with agent-driven steps for the routine path: data reconciliation, document drafting, and exception routing. A reviewer console for the edge cases the team actually needs to see.
What changed
Per-account ops time dropped ~85%. Billing disputes traceable to onboarding errors dropped to near zero. Leadership got a single source of truth on every new account.
Per-account ops time
~85% lower
Billing disputes
Near zero
First system live
~6 weeks
Headcount required
No new hires
What you actually get
Three things, done well.
01
Internal systems that run the operation
Custom internal tools, admin platforms, and workflow systems shaped around how your team actually works. The kind that replace the spreadsheet, not document it.
02
Operational AI agents in production
Agents that execute multi-step workflows — built on our open-source laravel-swarm framework. Auditable, reviewable, and tied to the metrics that matter.
03
Training so the team actually uses it
AI for the whole org, then department-by-department working sessions that turn the tools into daily practice. The most expensive AI is the kind nobody adopts.
How we work
Senior team. Direct access. No layers.
You work with the people building the system. Engagements scope tight, ship in weeks not quarters, and stay accountable to a real operating metric.
Full process →Week 1
Find the bottleneck
Sit with the team doing the work. Pick the workflow whose fix changes a real number.
Weeks 2–4
Ship the first system
Build the smallest end-to-end version that handles the common path in production.
Weeks 5–8
Cover the exceptions
Add the edge cases, the review console, and the agent steps that take the most painful work off the team.
Ongoing
Train the team, then exit
Department-level training, runbooks, and a clean handoff. We are not a permanent line item.
What we ship
The framework, in the open.
Anyone can claim AI competence. Few can point to the package powering it. Swarm is ours — a Laravel-native framework for agentic work, free, on GitHub, used on the agent builds we deliver.
Swarm — agent orchestration for Laravel
Multi-agent orchestration as a Laravel primitive. Define a swarm once, run it sync, queued, streamed, or as a durable workflow. Native to your stack. Auditable end-to-end.
composer require builtbyberry/laravel-swarmWhat's next
Swarm Cloud — the hosted runtime, soon. Ashlar — the operator-facing workflow product, soon.
Where to start
Three ways in.
You have a real problem
Scope a project.
Manual work, broken handoffs, or unadopted AI is already costing you. Bring the bottleneck — we will tell you whether it is a fit, what we would build, and what it would cost.
- — Direct conversation with Daniel.
- — No deck, no SDR, no pipeline drip.
You are exploring
Book a readiness call.
30 focused minutes on where AI would actually move your operation. Useful even if you do not engage us afterward.
- — Honest read on whether you are ready.
- — What would have to be true first.
You want to self-serve
Run the assessment.
Score one of your workflows on the five-layer Readiness Stack — Data, Process, People, Tools, Decision. Get a verdict and the weakest link to fix first. About three minutes.
- — Verdict: not yet, pilot, or build it.
- — Names the weakest link for you.
Tracking the products? Join a waitlist. Just curious? Read the notes or try laravel-swarm on GitHub.
Operational AI Readiness Call
30 focused minutes on where AI would actually move your operation.
Bring the workflow you would automate, the team that would use it, and the cost of getting it wrong. We will tell you whether it is a real fit, what would have to be in place first, and whether Built by Berry is the right partner — or not.
- — Direct conversation with Daniel, not an SDR.
- — No deck. No pitch. No follow-up sequence if it is not a fit.
- — Useful even if you do not engage us afterward.