Products
MCPAI Code ReviewDeveloper Q&A
Outcomes
Improve QualitySave tokensReduce rework
SecurityBlogDocsPricing
Book a DemoGet Started



Products

MCP
Improve AI coding agent output by connecting knowledge from your codebase, docs, and messaging platforms.
AI Code Review
High-signal PR feedback informed by how your system actually works. No noise or nits.
Developer Q&A
Fast, high quality answers without digging or disrupting coworkers.
Outcomes

Improve Quality
With plans and code that reflect how the system actually works you're one step closer to production.
Save tokens
Give agents the right context up front to reduce costly retrieval loops and tool calls.
Reduce rework
Get more accurate output on the first pass and spend less time reprompting and reviewing.
SecurityBlogDocsPricing
Log InBook a DemoGet Started

<
All posts
Perspectives
“Fake it till you make it” does not work for Developer Tools
Dennis Pilarinos
·
October 6, 2025

One of the most interesting interaction patterns we’ve seen at Unblocked is that every single new user, without exception, starts in the same way.

The very first question they ask is never a real question, it is always something they already know the answer to.

I suspect this is because most engineers are cautious and somewhat skeptical by nature. They tend to be dubious of new tools and they’ve been burned before by products that overpromise and under deliver (especially given the hype around AI developer tooling).

So the first interaction with Unblocked is a test. They want to know right away whether this system is worth their time and asking a “known” question is the simplest way to do that.

References as a “Trust Signal”

Correspondingly in those early moments, users don’t just look at the answer itself, they want to see the proof.

Our metrics show us that early in the user adoption cycle, we see a high volume of clicks to the references Unblocked provides with its answers. Users are validating not just what the system says, but where it came from.

What’s fascinating is that this behavior changes fairly rapidly. After a few days of consistent use, those clicks drop off and users stop double-checking and start trusting Unblocked’s responses. The references are still there, but they become a safety net rather than a crutch.

This transition tells us the product has crossed an invisible line, moving from “tire kicking” to being relied upon.

You only get one shot

This behaviour highlights something critical about designing developer products like Unblocked. First impressions really count and there is a very small window to establish trust.

If we nail that very first answer, we’ve earned the right to keep going. We passed the “BS test” and the user starts to form a habit (which is another blog post into itself).

However, if the answer is off, incomplete, or confusing, we begin in an (almost insurmountable) deficit. It takes far more effort to recover from that failure than it does to get it right the first time.

For us, that means we can’t think of “trust” as a nice-to-have or something that accrues over months. The very first interaction sets the tone for everything that follows.

Trust is critical for developer adoption

This insight has shaped more than just how we think about references or answer quality. It’s changed how we think about onboarding, product messaging, and even our roadmap. Lose a developer’s trust early, and you won’t get the chance to show the more powerful features that come later.

When you only have one shot to make a first impression, the right product decision isn’t always the most complex or technically impressive one. Sometimes it’s the one that simply proves: you can trust this.

Read More

March 24, 2026

•

Perspectives

The context problem moved from people to agents
Developers ship faster when they understand how their system actually works, not just how to write code. That context is nearly impossible for agents to reconstruct, so AI-native software teams end up wasting time and tokens fixing their code.

March 30, 2026

•

Perspectives

REPL is dead, long live REPL
Developers have always moved up abstraction layers: we stopped writing assembly so we could think in functions. Now agents are forcing us up another layer, towards intent but we’ve never had a probabilistic layer until now.
Get answers wherever you work
Book a Demo
vscode logo
VS Code
IntelliJ logo
JetBrains IDEs
Unblocked logo icon
macOS App
Slack logo
Slack
web icon
Web
Product
Get StartedBook a DemoDownload UnblockedPricingSecurity
Outcomes
Improve QualitySave TokensReduce Rework
Resources
BlogDocumentationPrivacy policyTerms of service
Company
About usCareersContact us