What 15 data-agent teams do about stale context, how Zep dates each fact, and 7 reasons context graphs work.
͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
|
|
|
|
|
|
|
NEW ON THE BLOG · 9 MIN READ
|
|
|
A data agent can write correct SQL and still pick the wrong table, metric or join key. We read 15 engineering write-ups, from teams at OpenAI, Anthropic, LinkedIn, Spotify and others, to see how they address it. Read more
|
|
By Patrick Kemp
|
|
|
|
KEY TAKEAWAYS
|
|
Data agents now fail on context more than on SQL. In a 2026 benchmark study, agents on the strongest model failed more often at finding the right data than at writing the query.
|
|
|
|
|
Most business context is outside table metadata. Decisions, exceptions and account terms are usually in documents and chat threads. In 13 of the 15 write-ups, people write or approve this context by hand.
|
|
|
|
|
A recent fact isn't always a valid fact. A message from today can describe a rule for last year's data. 10 of the 15 teams check for stale data or context, but none of the write-ups describes a record of when a business fact applied.
|
|
|
|
|
|
|
ALSO ON THE BLOG
|
|
|
|
|
|
Each fact in a Zep Context Graph keeps two timelines: when the fact was true, and when the graph learned it.
|
|
OCT 29 · 7 MIN
|
|
|
|
|
|
Why a graph of entities and dated facts gives agents better context than text chunks.
|
|
OCT 22 · 8 MIN
|
|
|
|
|
|
|
|
|
|
