Your agents can read code. Unblocked gives them the rest: synthesized context from docs, conversations, tickets, PR history and runtime signals that helps them generate reliable code in less time using fewer tokens.
External link for Unblocked
Unblocked finds answers from your team's code, docs, discussions and more. It augments your source code with related knowledge across the tools your team uses to build software, like GitHub, Slack, Confluence, Linear, Jira, and more. Now your team can get answers without having to interrupt a teammate. With Unblocked, your team gets: * Instant, helpful answers that explain (and visualize) the nuances in your codebase * A virtual team member to automate internal support directly in Slack channels * MCP server that brings your team's knowledge to supercharge tools like Cursor, Claude Code, and Copilot. * PR failure agent that diagnoses the root causes of CI failures and posts the fix directly in your Pull Request. Designed to be secure from day one, Unblocked ensures that your data stays private and isolated. Answers are permission-aware, which means your team only gets answers based on the data they're authorized to see. https://getunblocked.com/
Vancouver, British Columbia V6B 2X5, CA
San Francisco, California 94114, US
Unblocked reposted this
The engineering team at Codat has gone all-in on AI-native software development. The DX benchmark for AI-authored code share is 20% – Codat is at 75%. But their priority isn’t just more AI-generated code; it’s ensuring that everyone on the team can ship AI-generated code that adheres to their quality standards without taxing the system. For that, they turned to Unblocked. https://lnkd.in/gu2xnA9V
The engineering team at Codat has gone all-in on AI-native software development. The DX benchmark for AI-authored code share is 20% – Codat is at 75%. But their priority isn’t just more AI-generated code; it’s ensuring that everyone on the team can ship AI-generated code that adheres to their quality standards without taxing the system. For that, they turned to Unblocked. https://lnkd.in/gu2xnA9V
Your AI agents know how to write code. What it doesn't know is the shape of your organization: how things get done, how decisions get made. That gap is why doom loops happen. A context engine skips that whole loop to get to the right answer faster. Here's Brandon Waselnuk at QCon London on why context is the missing layer as you scale AI adoption in engineering organizations.
More context does not mean better output. AI agents need the right information in their context window, and nothing more. As soon as they go beyond 40% usage, the reasoning capability actively decays into the 'Dumb Zone'. Naive RAG can't tell the difference between the context that matter versus don't. That's a context engine's job. Here is Brandon Waselnuk unpacking why RAG isn't enough when teams are solving for context.
Your source code says one thing. Your Confluence doc says something different. Slack thread had a principal engineer explains a related issue with that approach. Which one does your agent trust? An agent without a context engine will find one and run with it. It won't reconcile the difference taking into account of who authored the doc, when it was updated, or what the source code actually does. A context engine does. It reconciles the different sources of information into the truth, respects who should have access to what, and delivers the right context at the right time. Here is Brandon Waselnuk sharing at Qcon London about what a context engine actually needs to do.
Our CEO Dennis Pilarinos sat down with Kevin Henrikson and Jason Shafton of Founder Mode to unpack what AI-native software development actually looks like in production. In our favorite clip, Dennis talks through the impact context has on agent-created code: https://lnkd.in/gnJfHA9q
This is Dennis Pilarinos. He built on the ground floor of Microsoft Azure. He was the first AWS employee in Vancouver. He founded buddybuild, a mobile CI/CD platform used by tens of thousands of developers, and sold it to Apple. Then he took a year and a half off, got his pilot’s license, and flew a small slow airplane with no destination and no return date. When he came back, he did not start a new company because of a market thesis. He started it because he missed the people. Dennis calls it “getting the band back together.” The team at Unblocked is basically the same crew from buddybuild. They wanted to work together again. The problem came second. The people came first. His rule for every career decision: 85 percent should be based on the people you work with. 10 percent on the problem. 5 percent on the position. The problem he picked is the one nobody talks about honestly. AI dev tools are solving the wrong problem. They focus on source code and ignore the context that actually makes code production-ready. The why behind the what. The Slack threads. The Jira tickets. The architecture decisions buried in a Notion doc nobody updated. Dennis ran a live experiment. He asked Claude Code to build a feature with no context. Then he ran the same request with Unblocked’s context engine pulling in historical conversations and internal documentation. The contextualized plan was dramatically better. The version without context proposed secrets management that would have taken down production services. His hiring philosophy is just as sharp. He asks one question in every interview: tell me about a really great day at the office and a really bad one. Then the follow up. What did you do to make sure the bad day did not happen again? If people are not taking control of their own environment, they will sit there and tell you what is broken instead of telling you what is broken and how they are going to fix it. He calls the people he hires “ambitious malcontents.” People who are frustrated with how the world works and would rather fix it than tolerate it. Curious. Humble. Full of user empathy. Those predispositions matter more than skills because skills can be learned. Predisposition cannot. His advice to founders: the path to your destination will change constantly. Get comfortable with massive uncertainty. And if you need something to keep you sane, find the thing that forces you to focus so completely that you cannot think about the business. For Dennis, that is flying a small airplane where the only job is not to kill yourself. The future belongs to founders who hire for ownership, build for context, and know when to take their hands off the keyboard and go fly. Link to the full episode in the comments.
Our very own Brandon Waselnuk shared at Qcon London earlier this year about one of the myths about managing context in agentic coding - the phenomenon known as 'satisfaction of search'. Agents stop searching once they found something that looks like the right answer, but might not be the right answer. This is one of the core reasons simply connecting a number of MCPs do not solve the context problem.
Our friends at DX just released a great study on the impact of AI on engineering velocity. Spoilers ahead: velocity gains aren't what you might expect. - Coding isn't the primary bottleneck — it only amounts to 14% of a dev's time - Time saved generating code is offset by increased reviews - AI adoption is stratified, making it harder to build shared workflows - Most developers are still learning, so gains are limited But here's the one we think matters most: - AI tools lack context — institutional knowledge is invisible to them The report puts a fine point on it: "An AI assistant can't reason over a Slack thread from an archived channel or the mental model of the engineer who built it." The model isn't the bottleneck. The context layer is. The report astutely advises teams to start with the basics: get your documentation in order, create your agents.md file, develop an onboarding guide. But we see these as core requirements, not accelerants. Teams ready to scale AI-native development require dynamic context, unified and reconciled from the places engineering decisions get made: Slack conversations, Jira tickets, Confluence docs, and PRs. That's the unlock for true engineering velocity — context that makes agents behave like teammates. For Unblocked customers, this is already the reality.
You just got paged... Error count is spiking in the payments processing service You open an incident and start digging through PRs, Slack threads, Jira tickets, and APM data to figure out what changed and why. Or, you can just ask Unblocked. Our new Datadog and Sentry integrations bring incidents, logs, and metrics into the Unblocked context engine, so you can ask one question and get a synthesized answer pulled from every relevant source. https://buff.ly/rkKPUdR
Whether it's written by an agent or engineer, the difference between good and great code has always been context. Today Unblocked is releasing our MCP in GA, alongside a set of tools that make it easier for agents to access the context they need to generate code that looks like it was written by your team's best engineer. See how it's already helping teams at Clio, HeyJobs, Tradeshift, and more: https://lnkd.in/ghgU6nGf