The Blackbird Roost | Issue #9 | October 2026

👋 Hi Reader, this is Bruno and Team Blackbird with a subscriber-only issue of The Blackbird Roost Newsletter.

Last month, we announced our Vancouver residency. This month, the calendar has caught up with us. Two workshops, two conferences, panels, demos and a three-minute pitch. Apparently, sharing a house was the easy part.

Before we pack, though, there's a release I've been waiting a long time to tell you about.

Blacklake can finally edit

Native editing in Blacklake is the most consequential release we're shipping this year.

Since launching a year ago, Blacklake has been the place where your content intelligence lives. It stores your translations, aligns them, diffs incoming changes and puts your assets and resources under granular control. It provides metadata and provenance about content history. It is the memory of your multilingual content operation.

But a memory you can't correct is only half useful, and for too long we have asked you to go somewhere else to fix a sentence. That was a gap, and we knew it. Today we close it.

You can now edit individual content units directly in Blacklake, changing only the passage that needs attention. For HTML content, the surrounding structure is preserved.

Edits are staged for review, then saved together as drafts. The live version stays in place: "Start your free trial" can remain live while "Start your 14-day free trial" waits for review. Birds use live content by default, and drafts only participate when you explicitly allow them.

Version history keeps the changes visible. A dedicated synchronization Bird can send drafts back to the connected system and store the finalized content in Blacklake, closing the loop.

The bigger picture is distributed editing with shared governance. An LSP can work in a TMS, an in-house reviewer in Blacklake, and a marketing team in its CMS. Blackbird orchestrates the handoffs; Blacklake preserves the content's history.

We built Blackbird to help content move without losing its context. Native editing brings us much closer to that promise.
​
​Explore native editing in Blacklake →

Video of the Month: Native editing in Blacklake

Prefer to see the release in action? Watch Dan Milczarski, our VP of Solutions, walk through native editing in Blacklake, then try it with a passage you've been meaning to fix.

​

Less waiting, faster Birds

Some product updates give you something new to do. This one gives you less time to wait.

One-minute polling is now available in Blackbird, enabling up to 5× faster event detection compared with five-minute polling. The maximum wait drops from five minutes to one. The average wait drops from two and a half minutes to thirty seconds. Your Bird can start sooner, and so can everything that follows.

Webhooks remain the fastest and most efficient way to trigger Birds. For integrations where webhooks aren't available, Blackbird checks for changes at regular intervals and triggers the Bird when it detects an event. That makes faster polling especially useful for AEM, Marketo, Sitecore, HubSpot, ServiceNow, Shopify, Magento (Adobe Commerce Cloud) and many cloud drive apps.

A small setting, with a noticeable effect on how quickly content moves.

What if your next software user is an agent?

Most language software assumes someone will log in, open a project and click through a process. What happens when an AI agent invokes those capabilities directly?

At the Localization Institute's Multilingual AI Roundtable in Vancouver, Mathijs Sonnemans, our VP of Product, and I will explore how MCP and headless architectures could reshape translation management, content operations and the language technology business itself.

MCP, the Model Context Protocol, gives AI assistants a standard way to discover and invoke capabilities that software exposes. Headless access lets those capabilities work without requiring someone to navigate an application's interface.

The opportunity is to combine content repositories, linguistic assets, translation engines, quality evaluation and workflow logic across the content stack. People describe what they need in natural language, and an assistant helps them query context or invoke governed Birds.

We'll bring live examples of what that looks like in practice, including how people who aren't solution architects can participate in designing and executing processes.

This also raises a commercial question. When agents become significant software consumers, how should we price value? Seats and connector fees will tell us less than execution and outcomes. Context, memory, evaluation, governance, security and reliability will matter even more.

Here's what that future of work looks like:
​

Should we change our process? Let's test it.

Marketing teams ask "Does A work better than B?" every day. They test headlines, landing pages and emails, and expect evidence. Localization can apply the same thinking to how we work.

That's the idea I'm taking to The Sandbox at TAUS Vancouver. Three minutes, no slides, and a proposal: make localization workflows testable.

At the strategic level, compare a human-led TMS process with an AI-led process built around orchestration, governed context and risk-based human review.

At the tactical level, compare LLM models, context settings or quality thresholds. Does raising the threshold for human review from 90 to 95 improve the result enough to justify the additional effort?

Here's a practical starting point. Take one content type from a CMS and process it through two paths. A follows the existing workflow. B uses AI with approved content, terminology and brand guidance from Blacklake.

Let B run in shadow mode, with its output kept out of production. Compare cost, turnaround time, editing effort and quality against the same criteria, ideally with reviewers who don't know which path produced each result. Then change one variable at a time and measure the tradeoff.

The winner may depend on the language, content and business objective. A product claim and a campaign headline may need different paths.

That's what we want to enable with Blackbird and Blacklake: orchestrate alternatives, provide governed context and capture evidence to guide decisions.

Start small. One content type. One language pair. One clear comparison. A useful answer to "Should we change our process?" might be "Let's test it."

The flock lands in Vancouver

We'll be exhibiting at LocWorld, but the booth is only one stop on the itinerary.

14 October: TAUS AIQE & Blackbird AI Orchestration Masterclass

Join Dan Milczarski, our VP of Solutions, Mathijs and me for a hands-on workshop on quality estimation and AI orchestration. We'll connect the operating-model discussion to practical workflow design.

​Explore the masterclass and register →

At the TAUS Conference: From TMS to AI Orchestration

Dan will join the panel on how translation management is evolving toward orchestration, and where automation and human expertise should meet. I'll also bring the experimentation proposal above to The Sandbox.

​See the TAUS program →

19 October: Multilingual AI Roundtable

Mathijs and I will lead a session on MCP and headless architectures at the Localization Institute's Multilingual AI Roundtable. Expect live examples and a discussion about what changes when language technology becomes accessible directly to agents.

​Explore the roundtable and register →

At LocWorld: A/B Testing the Localization Operating Model

Dan will join Erik Vogt, Ari Karampour of Indeed and Amir Kamran of TAUS to discuss how teams can compare operating models using measurable outcomes.

​View the LocWorld session →

If you'll be in Vancouver, reply to this email. We'd love to find time for a conversation, a demo or a coffee between sessions.

Thanks for being part of the Blackbird community. See you in Vancouver, or back at the Roost in November!

Bruno & Team Blackbird

[Website][LinkedIn][YouTube]