Cold takes | Penguin Strategies

Context Home: the screen every HubSpot agent reads first

Written by Perry Nalevka | Sep 16, 2026, 1:36:48 PM

"The AI got it wrong."

You'll hear that sentence a lot this quarter. HubSpot's own answer to it, delivered to partners before launch, is blunt: nine times out of ten, the AI was right about bad data, or about context nobody gave it.

Until this week that was an argument. Now it's a screen you can open.

What Context Home is

Context Home is a single view of what HubSpot currently believes about five things. Your business: brand kit, identity, location and scale, market, technology stack, products and services. Your customers: ICPs and personas, each editable individually. Your team and processes: user profiles and email personalities, sales methodology. Custom context: free text or uploaded files for the nuance that fits nowhere else, each with a visibility setting. Knowledge vaults: bounded sets of reference material attached to specific projects or agents.

It's under Agents → Agent Hub → Context, on every plan, and it replaces the old AI Data Sources settings page. It launches pre-populated from what HubSpot already knows (a crawl of your site, CRM-derived signals, files you've uploaded) and keeps recommending updates over time.

HubSpot's name for all of this is Growth Context. The distinction they draw is worth keeping: data is what happened (a logged call, a closed deal). Context is why it happened and what to do next (the email thread where the deal nearly died, the call where a customer explained why they buy). Agents need the second kind.

Why it matters more than any agent announced this week

Every agent in the Fall release reads from Context Home before it acts. Campaign Agent starts a plan already knowing your brand, ICP and past campaign performance. Content Agent loads your voice and competitive positioning before writing. Nurture Agent draws on it at send time. Deal Progression pulls the active sales methodology. Anything you build in Agent Builder inherits it by default.

So the quality of every agent's output is capped by the quality of this one screen. In most portals I've opened this year, that screen would be half empty and half wrong. An ICP from two years ago. Products that no longer exist. Qualification criteria that live in one rep's head.

Two mechanics that change how you work with it

First, edits are locked. Context Home reconciles values from several automated sources, but once a person explicitly edits a field, automated updates won't overwrite it. Whatever you fix stays fixed.

Second, uploads produce suggestions, not writes. Drop in a 40-page sales methodology and the extraction pipeline pulls out deal stages and qualification criteria and surfaces them for review. Nothing appears until someone approves it. If you upload a document and "nothing happens," that's the design: it's waiting for a human.

Permissions are role-based only. Super Admins and partner admin roles can create, edit and delete; everyone else can view; any user can manage their own email personalities. There's no object-level granularity.

Context versus knowledge vaults

These get confused constantly. Context is foundational and ambient: it applies automatically across every AI tool in the portal. A knowledge vault is deliberate and bounded: a set of files or CRM segments attached to specific agents or projects. Up to 50 vaults per account, files to 50 MB each, images searchable alongside text. Put your methodology in context. Put the product spec sheets a support agent needs in a vault. Mix them up and you end up with the same PDF uploaded eleven times and no idea which agent reads which.

How to audit yours in an hour

Open Context Home and read it as if you were a new hire on day one.

Business tab: is the description accurate? Are the products listed the ones you sell now? Is the competitive landscape current? Customers tab: is the ICP who you sell to today? Do the personas match the people on your last ten calls? Team and processes: do the email personalities sound like your best rep? Is the sales methodology the one your pipeline actually uses? Custom: what's missing that a new hire would need to know in week one?

Note which areas are populated, stale or empty, and when each was last reviewed. Then open Agent Hub next door and answer three questions: which agents are active, who turned them on, and what has each produced in the last 30 days.

Do the Business tab first. It's usually the weakest and the fastest to fix, because the information already exists in someone's head or a slide deck, and the improvement in agent output is visible immediately. Customer data cleanup matters more in the long run and takes far longer. Start where the win is fast.

The mistake to avoid

Treating this as a one-time setup. Products change, positioning shifts, reps leave. Context decays. Put a quarterly re-read on the calendar, and make one person the owner.


Part of our guide to the HubSpot Unbound 2026 release: everything HubSpot announced, on one page.