AI only works with context. Today we're shipping the Enriched Context Stack — the release where years of foundational work becomes visible in the product.
AI only works with context.
We started Chord long before the agentic commerce category had a name. The operators we've spent years working alongside were drowning in fragmented systems, and it was already clear that AI alone wasn't going to fix it. What was missing was a real foundation underneath, one that could give AI the context it needed to actually run a commerce business, not just analyze one.
“Chord is the context layer for agentic commerce. AI only works with context — and that's the bet we've been making, in the trenches, for years.”
The frontier labs and the analysts are arriving at the same place. OpenAI's recent writeup of their in-house data agent details six layers of context built specifically to keep strong models from misinterpreting their own data. a16z's Your Data Agents Need Context argues that a modern context layer is a superset of the semantic layer. I loved reading about how giants in the space arrived at the exact gap we hit in our first year and have been closing ever since. Both are worth reading. Both reinforce what we've been heads-down building.
Today we're shipping the Enriched Context Stack — the release where that work becomes visible in the product.
What's new
The Enriched Context Stack is four upgrades to Chord AI Copilot, designed to reinforce each other.
Memory (or more memory)
Most AI systems don't learn from use. With this Memory release, Copilot does. Corrections, repeated nuances, the rules it picks up while answering questions — that knowledge carries forward. Real work is cumulative. The systems we rely on should be too.
Runtime Context
When Copilot encounters something unfamiliar — a misspelled product name, an unknown variant, a table it hasn't seen before — it no longer fails quietly. It investigates in real time, querying the warehouse to inspect schemas and resolve unknowns before answering. Those discoveries feed back into Memory, so the same question is faster the next time.
Better question interpretation
Copilot now does a better job understanding what you're asking before it tries to answer. It identifies intent, decides when clarification is needed, and shows its work along the way. Less friction at the start of every interaction. Better outcomes at the end.
Self-evaluation
Answering a question is half the job. Knowing whether the answer is reliable is the other half. Copilot now checks its own work after generating a response, and if something looks off, it investigates further or asks a follow-up rather than handing you a confident wrong answer.
Together, these changes deliver measurable gains in the three properties that earn trust in AI for commerce ops: accuracy, reliability, and relevance.
Why this is different
Most platforms separate the data foundation from the intelligence layer. Data sits in one system, gets analyzed in another, and gets handed off to AI sitting on top. Every handoff introduces gaps. Those gaps show up as inconsistency, latency, and answers your team can't trust. They can be nearly impossible to audit when things go wrong, and don't get me started on governance.
Chord is built differently. The data foundation and the intelligence layer are designed together, from the start. That's what lets Copilot operate with complete context, not partial context stitched together at runtime. When context is incomplete, outputs are unreliable. When context is coherent, the system can actually reason.
That architectural difference is what determines whether AI works in practice, not in a demo.
What's next
The Enriched Context Stack is the foundation. What we're building on top of it is agentic ops.
There is a lot of energy in the industry right now around agents, and deservedly so. Clever, capable, increasingly autonomous agents finally seem within realistic reach. We're excited about our agents, too! But, and this is a big one, they (the agents) can sometimes seem more capable than they actually are. After all, an agent is only as good as the context it operates on, and an agent without a foundation of trust will never be capable of truly doing the job without a ton of oversight (and that's not the future we all want). The cleverness of the agent doesn't change that. The context underneath it does.
That's why we built this in this order. Trust first. Agents next. We're excited to start testing ours against a foundation we know we can rely on, and to share what we learn as we go.
The teams who win in the next era of commerce won't be the ones racing to layer agents onto broken foundations. They'll be the ones who fix context at the root, and let agents operate from there.
That's the work. We're grateful at Chord to be doing it with you.