May 5, 2026
47 steps to nowhere
Show HN: Airbyte Agents – context for agents across multiple data sources
Airbyte says it can stop AI from getting lost, and the comments are already buzzing
TLDR: Airbyte launched a tool that tries to help AI assistants find the right company info faster and make fewer mistakes across multiple business apps. Commenters were mostly impressed, but the big debate was whether the company’s self-run benchmark can be trusted, even with the test code made public.
Airbyte just rolled out a new tool meant to help AI assistants pull info from lots of workplace apps without wandering into a 47-step disaster. CEO Michel pitched it with a painfully relatable horror story: an AI asked a simple business question, took forever, burned through loads of computing effort, and still got the answer wrong. Airbyte’s big promise is basically, what if the bot already had the important context ready, instead of frantically rummaging through Slack, Salesforce, Zendesk, and everything else in real time.
And the community reaction? A spicy mix of genuine hype, nerdy curiosity, and polite side-eye. One commenter immediately zoomed in on the real juicy question: are some AI models just better at finding their way through messy webs of company data than others? Another Airbyte team member jumped into the thread with a classic startup-defense move: yes, we know benchmarking your own product looks suspicious, and yes, that’s why they published the test setup for everyone to inspect. That gave the launch an undercurrent of "trust us... but also verify us" energy.
The vibes got warmer when former employees and users piled on with congratulations, praise, and a little inside-baseball gossip. One user casually dropped that Airbyte’s old white-label option had been deprecated, which added a tiny whiff of product-drama to the celebration. The funniest flex came from swyx, who cheered the company for "adapting to the AI age" and "using the crap out of Devin," turning the comment section into a mini reunion mixed with a future-of-work meme. In other words: less fighting than usual, but plenty of startup intrigue.
Key Points
- •Airbyte launched Airbyte Agents as a unified data layer for AI agents working across multiple operational systems.
- •The company says direct API usage and thin MCP wrappers leave agents handling authentication, pagination, schema differences, and entity matching across tools.
- •Airbyte Agents is built around a Context Store, described as a data index optimized for agentic search and populated by Airbyte’s replication connectors.
- •Airbyte says the product was motivated by an internal 47-step agent workflow that was slow and returned an incorrect answer.
- •In Airbyte’s published benchmark, its Agent MCP used up to 80% fewer tokens for Gong, 90% fewer for Zendesk, 75% fewer for Linear, and 16% fewer for Salesforce than compared MCP approaches.