Meta bought Stilla.ai, a Stockholm AI startup, on 9 September, folding it straight into Meta Business Agent, the chatbot already running customer chat for over a million businesses on WhatsApp, Messenger and Instagram. If you run any automated WhatsApp flow, this is the acquisition worth reading about this week. Not for the name, you won't have heard of it. For the problem it actually solves, keeping context straight when an AI agent works across real business systems.
What Is Meta Business Agent, and Why Did Meta Buy Stilla?#
Here's what's actually confirmed. Stilla launched eight months ago, in January 2026, after raising $5 million in pre-seed funding led by General Catalyst. It built what its founders call a shared context layer, tying together tools like Slack, GitHub, Linear and Notion so a human and an AI agent could work off the same live memory instead of five disconnected app windows.
Axios first reported the deal on 9 September, three days after Meta's Muse agent launched with its own WhatsApp access. Meta's stated reason is blunt, accelerate Meta Business Agent, the tool already handling customer service and transactions for over a million businesses on WhatsApp, Messenger and Instagram. Neither company has disclosed a price. Stilla's own announcement says the product keeps running as-is for existing customers, no forced migration, at least for now.
Does This Change Your WhatsApp Chatbot Today?#
Not yet. If you're not running any kind of WhatsApp automation, or you're still handling under maybe 50 conversations a day through a human inbox, this doesn't touch you. Keep reading anyway, because it's coming.
If you are running Meta Business Agent, or something on the WhatsApp Cloud API for lead qualification, appointment booking or order status, this is the direction your tooling is headed. The gap Stilla was built to close is what people in this space call the messy middle, an AI agent that forgets what a customer said two messages ago, or one with more access than it should have, taking an action nobody approved.
Most SMB chatbot builds, ours included in earlier versions, handle this with brittle workarounds instead of a real fix:
- Session variables that reset when a conversation goes quiet for too long
- Manual handoff rules to pull a human in when the bot loses the thread
- A hard reset of the flow when it gets confused, which the customer feels as the bot "forgetting" them
If Meta ships proper shared memory into Business Agent, that's fewer of these to maintain. My read is this matters most for GCC and Indian businesses running high WhatsApp volume, and a lot less for a US or Canadian SaaS company doing occasional support chat on the same channel.
The Real Trade-offs, Not Just the Upside#
The upside, if it ships as described, is less duct tape. A shared context layer built into the platform you already use for customer chat means fewer separate memory systems to sync. It could also mean better handling of multi-step conversations, the kind where a customer asks about pricing, disappears for two days, then comes back expecting the bot to remember.
The honest downside is we don't know the timeline. Acquisitions like this take months to land in a shipped product, and Meta has a mixed record on how fast acquired tech reaches third-party developers versus staying internal-only. There's also an unanswered data question, whose shared memory is this, and what happens to it if you switch providers later?
More of your customer relationship sitting inside Meta's stack is a concentration risk, not a free upgrade. We got caught off guard by the WhatsApp API pricing change hitting October 1 with less warning than we'd have liked, and a platform holding more of your context can move on you the same way.
How We're Already Adjusting Our Builds#
At WebEpex we build WhatsApp and chatbot automation for GCC and Indian SMBs on n8n and Voiceflow rather than betting on a single platform's roadmap, and this doesn't change that. It just means we're watching Meta's developer docs more closely, which lines up with why we already flagged checking your Meta Ads AI connector permissions a few days ago. Same platform, quietly widening what its AI layer can touch.
We reviewed the announcement the day it broke and checked it against two live builds. On a communications stack we scoped for a client running mutual fund and insurance sales, we already locked in n8n for the context-passing piece instead of waiting on a platform to solve it for us, alongside AiSensy for WhatsApp, one company number, and after-hours calls deflecting straight to WhatsApp. That's the manual version of what Meta is now buying its way into.
On a Voiceflow build we shipped for a menswear client, the same context problem showed up as a write-and-read asymmetry in Voiceflow's MCP layer. We ended up delivering rich content as paste-in blocks instead of live carousel pushes to work around it. Different platform, same underlying issue. Every builder right now is solving persistent memory by hand.
Our stance this quarter is to keep building that layer ourselves rather than betting a client's whole flow on Meta shipping this well. If Business Agent's memory turns out reliable and exposed through the API, we'll migrate pieces in. Until it's in the developer docs, not the press release, we're not rebuilding around a promise.
What I'd Tell a Client Asking About This#
Do nothing to your current WhatsApp setup today. There's no API change yet, nothing is deprecated, nothing breaks. What I would do is bookmark Meta's developer changelog and check it monthly instead of daily, because acquisitions like this take quarters to surface as an actual feature. If you're mid-build on a new chatbot right now, ask your developer how they're handling conversation memory today, not whether they've heard of Stilla. That answer tells you more about your build's quality than any acquisition news will.
We've written up the mechanics of a proper handoff before, in our WhatsApp lead response build, if you want to see what "reliable memory" should actually look like before Meta ships its version.
If you want a straight read on whether your WhatsApp setup is actually built to last through changes like this, send me what you're running and I'll tell you in two minutes, no pitch attached. cal.com/webepex/growth-review
