2389 Research
The Factory
AI enablement for serious engineering organizations.
Build the AI-enabled engineering system your organization actually needs.
Writing code is getting cheap. The expensive parts of engineering now are judgment and context. We help you design and prototype the code agents and AI infrastructure your team keeps wishing for — built against your real systems, budgets, policies, and data boundaries.
Our point of view
The future of engineering is multiplayer.
Most organizations respond to AI by giving each engineer a coding assistant. That makes a few engineers faster and leaves the organization unchanged. The change that matters happens when the whole team shares context, practices, and infrastructure.
AI should expand capacity and ambition.
The systems we design create cost-optimized engineering capacity at scale. We call it tokenoptimization, not tokenmaxxing. Doing today's work with fewer people is the least interesting option. Spend the new capacity on work that never made the cut: the migration nobody could staff, the tool everyone wanted.
The why must remain human.
Agents can carry the mechanics; people orchestrate them. Agents can't be accountable, and they don't care why the work matters. Purpose and judgment stay with people, and the tooling should leave people with more agency.
Your constraints are design inputs.
Security review, privacy rules, audit trails, procurement, where your code and context travel: requirements to design for, not obstacles to route around. Trusted intermediaries in front of the labs; proof that nothing phones home. The interface between the models and your institution should belong to you — including, when it earns its cost, a model tuned to your work.
Workflows matter more than tools.
Any tool you adopt is temporary. The way your team works with agents is what compounds: how work gets split, where context lives, how agent output gets reviewed.
The mechanics will change. The values should not.
Build something bounded, use it, let evidence drive the next decision. Your standards for trust and judgment should outlast the tooling.
What we build together
Built against your real systems.
- Coding-agent practices that hold up across teams, not just for one power user
- Agent interfaces wired into your repositories, permissions, and workflows
- Systems that capture organizational context and keep it where agents can find it
- Patterns for review, testing, and oversight when agents write most of the code
- Cost controls that keep agent runs token- and budget-aware
- Experiments that settle what to build, buy, govern, or own
How we work together
There's no fixed on-ramp.
A day at our Chicago lab
Come to Chicago. See the systems we run, and pressure-test yours against them.
A working session with your technical leads
Your leads, our lab, one hard problem on the table.
Hands-on time with one of your teams
We work alongside your engineers in their real repos and workflows.
A bounded prototype
A scoped build that settles a question with evidence instead of slideware.
A research question
Something you need answered before you commit budget. We go find out.
Every format has the same goal: your own evidence, your own capability, and the ability to keep going without us. If you're shopping for staff augmentation, model licenses, or a big-firm transformation program, we're the wrong shop.
Why 2389
We hit these questions in our own work first. Our lab builds agent infrastructure, coding systems, orchestration tools, and context-management software. We learn by shipping, we publish what we find, and we open-source what we can.
Questions
What organizations ask us
What does an engagement look like?
There's no fixed on-ramp. Start with a day at our Chicago lab, a working session with your technical leads, hands-on time with one of your teams, a bounded prototype, or a research question. Every format ends the same way: your own evidence and the ability to keep going without us.
Who is this for?
Engineering organizations with real constraints: security review, procurement, audit trails, systems that predate everyone in the room. If AI adoption were easy where you are, you wouldn't need us.
Is this staff augmentation?
No. If you're shopping for staff augmentation, model licenses, or a big-firm transformation program, we're the wrong shop. We build capability that stays when we leave.
What do we own when it's done?
Everything. The prototypes, the practices, the evidence. The interface between the models and your institution should belong to you — that's the point.
What about our security and data boundaries?
They're design inputs, not obstacles. Trusted intermediaries in front of the labs; proof that nothing phones home. We design for your constraints because that's what makes the result usable.
What does it cost?
Depends on the format. A lab day and a bounded prototype price differently. We'd rather run a small experiment that settles a question than a big program that doesn't.
Compare notes
Let's compare notes.
Bring us the problem that makes AI adoption hard where you are: the mission, the backlog, the policy, the systems mess. Build more. Keep the why human.
No spam.