An update to The Best Context Isn't the Most Context. The example discussed below — an internal customer.io tool called Chiefys — was described by Matthew Wensing (VP Product & Design, customer.io) in conversation with host Aakash Gupta on The Growth Podcast, "How a VP of Product Uses Claude Without Producing Slop" (YouTube, published June 9, 2026). Full citation at the foot of this piece.
A note on where this comes from. I have no relationship with customer.io — financial, employment, or otherwise. I have not used Chiefys and have never seen it run; everything below is drawn from Wensing's public description of it, and my reading should be weighed in that light. I am writing because the tool directly operationalizes an argument I had made, and encountering that was reason enough to respond. Read it as one author reacting to a public account of someone else's work, not as a verified case study.
When I argued that context is a product and that deprecation has to be a first-class operation, the honest weakness of the piece was that it described a discipline almost nobody had built yet. The principle was right; the example was hypothetical. It is not anymore. A company operating at real scale has built the mechanism the essay said was missing, and it is worth walking through what they got right — because it confirms the argument and sharpens the part most teams will still skip.
The tool is an internal Slack bot at customer.io called Chiefys. By Wensing's account it was built by the company's co-founder and CEO, Colin Nederkoorn, and Wensing noted he felt free to describe it because Nederkoorn had already discussed it publicly on stage. What follows is my reading of Wensing's description, not a report on the system itself. It does two things.
It treats a document set as ratified, not just present
First, when anyone produces a new document — an analysis, a pricing rationale — they can run it against Chiefys, which holds a curated set of twenty to fifty company documents that are "the gold standard, the ratified, verified documents that we operate on." Chiefys checks the new artifact for discrepancies against that operating model and flags where it conflicts.
Read that against the original claim: context is a product, not a pipe. This is what owning the product looks like. Nobody piped the entire company wiki into Chiefys. Someone made an editorial decision about which twenty to fifty documents are true — which ones define how the company actually operates — and everything else was left out. That curated set is exactly the "right altitude" the essay was after: enough to govern a decision well, not the maximum the window could hold. The size is the point. A bot checking against five thousand documents would be checking against the noise; a bot checking against fifty ratified ones is checking against the company.
It makes deprecation a first-class operation
The second use is the one I want to dwell on, because it's the half teams reliably skip. When the company ships something new — a strategy shift, a launch — the existing documents that described the old reality are now wrong. Chiefys audits them. It surfaces the contradiction directly: eight of these twelve documents disagree with this. You're recency-biased and haven't noticed. Is that intentional?
That is the operation the essay called for and almost no org funds. Adding context feels like progress; deleting it feels like loss, so stale documents accumulate as landmines an agent will step on at the worst moment. Chiefys inverts the default. It doesn't wait for someone to remember that the pricing changed in March — it actively finds the dead definitions still sitting beside the live ones and forces a decision: retire it, or declare on the record that the contradiction is intentional. Silence stops meaning "keep forever." Someone has to say "still true" or the conflict gets raised. That's the freshness contract and the tombstone, made operational, running on a schedule, instead of living in a governance doc nobody reads.
And note why the team built it, as Wensing tells it. Not as housekeeping for its own sake — as a direct answer to the question the essay ended on. "No one has time to go back and look at all those," he says, "and that tends to be why Notion gets stale." The reason context rots is that auditing past artifacts is unrewarded work that humans deprioritize forever. Chiefys is the org deciding to reward subtraction by automating it — the muscle the original piece said was the competitive skill.
What it sharpens about the original argument
Two refinements come out of seeing this in the wild.
The first: deprecation is easier to fund as consistency-checking than as deletion. I framed the discipline around removal — give context an expiry, retire the superseded version, reward the team that pruned two hundred stale docs. That's correct but it sells badly, because deletion reads as loss. Chiefys reframes the same operation as "find where our documents disagree with each other," which reads as quality, not subtraction. Same outcome — stale context gets caught and retired — but a framing leadership will actually pay for. If you're trying to get this funded, lead with contradiction-detection, not pruning.
The second: the ratified set and the deprecation engine are the same product, not two. The gold-standard set is only trustworthy because something keeps auditing it against new reality. A curated set with no deprecation mechanism is just a slower-rotting pile. The essay treated "right-sizing" and "old context is worse than no context" as separate sections; Chiefys shows they're one loop. You earn the small, trusted set by continuously checking it, and the checking is what keeps it small and trusted.
What to do now, made concrete
The original "what to start doing now" still holds — inventory context, right-size it, name an owner, govern at the context layer, build the muscle of subtraction. Chiefys just gives you the minimum viable version of all five in one artifact you can actually build this quarter.
You don't need an agent platform. You need a defined set of documents you're willing to call true, and a scheduled check that flags every new thing against them and every old thing they now contradict. Start with one decision domain — pricing, or your operating model — and fifty documents, not five thousand. The bot is downstream. The work is the editorial judgment about which fifty, and the discipline to let the check retire the rest. That was always the real shift: not access, but deciding on purpose what deserves to be in the room. Chiefys — as customer.io's team describes it — is what it looks like when an org stops planning to do that and starts.
Source. Matthew Wensing (VP Product & Design, customer.io), interviewed by Aakash Gupta on The Growth Podcast, "How a VP of Product Uses Claude Without Producing Slop," YouTube, published June 9, 2026: youtube.com/watch?v=yDeFGKaSoX8. Chiefys is an internal customer.io tool attributed in that conversation to the company's co-founder and CEO, Colin Nederkoorn. All descriptions of it here are drawn from Wensing's on-air account, not from direct use or any company-provided material; quotations are transcribed from the episode and lightly punctuated for readability. I have no relationship with customer.io, its employees, or the podcast.