Wade Evanhoff

Having simple E2E tests goes a long way in this new world of rapid development with AI. It certainly makes me feel better with a second pair of eyes on each deployment.

At Caldera we're rolling this out across our internal web projects.

You can
- Provide target devices to emulate (iPhone 15, etc.)
- Provide your LLM model ( with API Key) of choice
- Specify headless or headed (you can see it clicking around and navigating in a browser)

It can do this locally, but also in your build pipelines to give peace of mind with more E2E coverage. Paired with Vercel's ephemeral preview environments, having an AI test engineers test the app in your preview environment before you even need to review the PR is a great thing.

This last weekend I built WALL-TE (Web Automated Large Language Test Engineer)("walt" for short), which is a CLI AI powered test engineer. You (or AI I suppose) writes natural language, plain text tests (no coding needed) and your LLM of choice will execute the test and validate assertions.

You might spend time writing a massive prompt full of context in the same time you probably could have gone and done the work yourself. The art is in the balance.

Adding the right context to JIRA tasks has been an art in just enough direction with reference files and examples without wasting time being overly prescriptive. It's the same thing with AI.

The New Skill in AI is Not Prompting, It's Context Engineering Context Engineering is the new skill in AI. It is about providing the right information and tools, in the right format, at the right time.

While the new term "Context Engineering" is a better definition of what we are doing, I can't help but think we are just re-inventing new terms for old ideas. I've been "context engineering" for 10+ years as a tech lead on projects, long before AI and LLMs.
www.philschmid.de/context-engi...

And all of this in spare hours between meetings or after kids go to bed, not because we have to but because it's fun to build and the value is so tangible.

6. We connect our IDEs to it with AI (Copilot, Cursor, etc.) via the MCP server so we are always building against the latest requirements.

We are actually using the platform itself to build out features for the platform.

4. Built with and for AI from the start. MCP is a first class citizen and it isn't just a wrapper around the API, it's purpose built for how Agents actually function.
5. We integrate our Slack conversations, transcription services, and email clients with it so requirements are never lost

Highlights:
1. Multi-Tenant / Organization
2. 1,000 + unit and integration tests
3. Temporal data with event sourcing, which means full audit history and versioning of all projects, requirements, proposals, users, etc.

For one, we haven't found a solution that fit our needs, and secondly, we now have the capacity with AI tooling to *own* our own platforms rather than pay per seat licensing fees for products we only use 1/10th of.

For an example of how AI is changing our industry, at Caldera, with spare hours, we've built our own AI-native, requirements management platform. It's kinda the equivalent of Cursor or Copilot but for product owners and project managers.

You’ll forget how to fish if you didn’t enjoy fishing that much in the first place.

You’ll keep fishing if that’s what you enjoy.

Teach a robot to fish if your joy is more about providing more fish to more people

Teach a robot to fish if it provides you more value and time for the things you love

~Vibe coding~ has affirmed one thing: the hard part of building a successful product has never been writing the code. My best solutions as an engineer are the ones that require no code. AI is very eager to write code, but I have yet to experience a scenario where it has told me it's not a good idea.

There’s always been some saying or belief that software engineers are “inherently lazy” and we tend to build software to automate away challenges. I see this person doing no different. My hopes is that this new wave of engineers becomes more attuned to the real problems that need solving.

If the foundational AI model companies are suggesting senior level programming AI agents in the next 5 years, some companies can afford to take a bet on that. Worse case, they have to dip into a saturated pool of engineers looking for a job.

I’ve seen this too! And even more, I’ve seen other roles on the fringes of engineering start producing technical value with AI. Platforms like N8N are being used to build internal automations and services by traditionally non-engineering roles, using AI for assistance.

I’d say AI empowers good engineers to produce value quickly enough to not warrant hiring the junior engineer in the first place. As someone responsible for managing & hiring engineers, this is happening now, it’s not a prediction.

I’ve been using Brave browser for web and mobile for years! That plus Kagi for search and I don’t even think about it. Works with all chrome extensions, dev tools are great, good default ad and third party blocking. Just squint past the crypto stuff and you’ll be 👌

It’s neat to see NotebookLM folded into a big product feature like this. Worked really well for my wrapped playlist and it’s a fun and unique addition to their existing AI experiences. Kudos!

"assume" wasn't the right word, I understand it isn't sentient, but it's more likely that a function "calculateFemaleSalary" is found in code calculating pay disparity and research vs an HR system. This isn't reflecting some AI bias that women deserve less pay, it's reflecting statistical evaluation

It seems you are directing the AI towards that answer by even proposing a different function for gender. Without context it could assume you are actually trying to calculate known pay disparities.

I wouldn’t consider that a proper example of bias but I’m certain the models do reflect our biases.

I use Brave and just disable and ignore the crypto.

To me, the only hang up with Blazor is the tooling. VS Code still continues to struggle with intellisense often for any moderately large Blazor application. Hot reload is picky and the whole thing can often be frustrating.

I come from a C# .Net background and I find Blazor to be the most intuitive component based framework I’ve used (compared to React, etc). It just works, and it implements concepts like server rendered components or WASM with very little configuration.

I love writing code but my thought and architectural leadership is significantly more valuable to the organization and product delivery process. It’s also hard to play both IC and tech lead without the competing priorities of properly defining work for others and delivering on my task. Both suffer.