Bumblebee Studio, led by Edo Ceder
A project often starts with whatever is already holding the work together: an old system, a spreadsheet or somebody who knows every exception by heart. We learn that setup first. Then we turn it into software and stay involved when it goes live.
Start with the messy version
Which file gets emailed around? Who approves what? What happens when the ordinary path breaks? A screen share with the people doing the job is usually more useful than a polished brief.
Write down the rules
Who may see or change the data? Which system supplies it? Which exceptions cost money or stop the work? These questions are cheaper before they become code.
Build a working path
We build enough of the interface, backend and integrations for someone to complete actual work. AI is included when it has a defined job—not sprinkled over the product.
Put it to work
We release carefully and watch people use the system with their own data. That is when the missing shortcut, odd exception or confusing label finally becomes obvious.
What we take on
We may be starting from scratch, adding AI to a product already in use or opening up an old system nobody wants to touch. Those are different jobs, and we treat them that way.
Ori coordinates a produce distributor’s daily operation; Colleads handles sensitive referrals between health professionals. Work like this includes the screens, data model, permissions, outside services and the awkward cases generic tools miss.
React · Next.js · Supabase · Bubble · APIs · WhatsApp · payments · browser testing
In Plenty, agents take part in a structured family-finance workflow. Receipt Finder is being built to match statements with documents in authenticated email. The model has a defined job, bounded tools and a way to hand control back to a person.
OpenAI · Claude · agent tools · RAG · MCP · structured output · cost and audit controls
Before replacing older software, we find the business rules it carries and protect current behavior with repeatable tests. Then we move the parts worth changing, one working flow at a time.
Lotus Domino · Java · DB2 · XML/XSL · system mapping · regression · phased migration
Client work
Plenty supports family-finance work over WhatsApp. Ori coordinates daily produce distribution. Colleads manages referrals between health professionals.
AI-native financial product
A WhatsApp-first family-finance platform that combines structured financial workflows with carefully bounded conversational agents.
Agricultural operations platform
A Hebrew B2B system coordinating daily produce availability, orders, allocation, fulfillment and reconciliation across growers, distributor and wholesalers.
Professional referral network
A mobile-first platform that helps health and mental-health professionals find suitable colleagues and manage sensitive client referrals.
Studio products
st4ck keeps requirements, code context and regression tests connected during product delivery. Receipt Finder uses bank and card statements to search authenticated email for matching invoices and receipts.
Product lifecycle workspace for humans and AI agents
st4ck connects stakeholder input, requirements, decisions, code context, engineering work and regression testing in one structured workspace.
Explore the product
AI-assisted document collection for bookkeeping
An agent workflow that reads bank and credit statements, identifies merchants and searches authenticated email for the invoices and receipts needed for tax reporting.
Explore the productWork in different settings
At Aladdin, Teva and Menora, a change could affect systems already in daily use. A smaller company often needs to remove one stubborn block in the work without turning it into a long program. A startup needs the first version to settle a question before it funds the next one.
The rules may be split between code, documents and people’s memory. We map them, add tests around the workflows that must keep running and change the system in stages.
A small team feels every repeated step. The software should remove enough work to justify the time and money spent building it.
Before building too much, we choose the one question the first version must settle. Then we put it in front of users.
Past and current work includes
Once it is live
Plenty is ongoing, and we still maintain Ori and Colleads. That is deliberate: real use uncovers the cases nobody remembered to mention, and the product is better when the people who built it are still around.
About Edo
I’m Edo Ceder. I have spent more than 30 years building software, including enterprise work with Aladdin, Teva and Menora, startup work and the products shown on this page.
I run the early sessions, make the architecture decisions, develop hands-on and review what ships. When a project needs a larger team, two developers and a project manager join me. Clients still work directly with me, and I remain responsible for the result.
No specification yet?
A screen share, a spreadsheet or the current product is enough to start. Tell us what is frustrating and what must not break. We will tell you what we think the next step should be.
Start a conversation