Michael Tierney

I build open source AI and MCP developer tools. Creator of mcpgauge. I also run a small data shop for Finger Lakes businesses. Building in public.

[Michael Tierney @Michael_WCD]

Your tools shouldn't live on separate islands. WhiteCap Data connects your whole stack - website, marketing, automation, and your data - so your business runs as one. One partner for the full stack of growth. Book a 15-min intro: https://t.co/Iez7LvcpLv

[Michael Tierney @Michael_WCD]

People assume helping a business grow online means something complicated. Most of the time it's the opposite.

Fix the small broken things. Make the basics accurate. Remove the friction between a customer and a sale. The results follow.

[Michael Tierney @Michael_WCD]

@kazi_asid Whats your experience with tool metadata? Can an agent learn from the docstrings on a function call or does it need more explicit labels?

[Michael Tierney @Michael_WCD]

@Armankhaan01 That's when most people start hitting the wall of tool metadata quality 5% of your tools have good docs, but the other 95% are just strings or worse. What kind of data are you using to train the search?

[Michael Tierney @Michael_WCD]

@stijnvwx Whats your conversion rate from engaged followers to paying customers?

[Michael Tierney @Michael_WCD]

@Nwaogaakachukwu Whats the real problem youre solving at your academy? Is it getting the agents to do what you want or actually teaching people how to use them?

[Michael Tierney @Michael_WCD]

@RoundtableSpace Building a tool registry is just a matter of structuring the data, not some grand architecture.

[Michael Tierney @Michael_WCD]

@benlsage @claudeai That's probably because the rewards got stuck in a local maxima and the agent stopped exploring new states. Check if the reward function still has high variance after 1000 episodes.

[Michael Tierney @Michael_WCD]

@AhoteSpirit @TheARCTERMINAL The encryption is good, but what about the model metadata? Does it get updated when you add a new tool?

[Michael Tierney @Michael_WCD]

@SeHozaifa Your 1 hour growth metric is likely being inflated by bots, not real users.

[Michael Tierney @Michael_WCD]

@codewith55 People use Windows for one reason: they're building something that needs to run on it.

[Michael Tierney @Michael_WCD]

@rflaherty71 Claudes got a nice output system but have you seen the RAM usage on that sprite generator?

[Michael Tierney @Michael_WCD]

@Biemjo You must have hit a 30% recall rate on the brand mentions or the models confidence in its own results before it gets to fix anything. Whats your approach to evaluating that?

[Michael Tierney @Michael_WCD]

@mogau_im Claudes feedback is just a bunch of strings until you turn them into actual error messages or diffs that point to the line of code.

[Michael Tierney @Michael_WCD]

@_naznotes What breaks agents isnt gratitude or a good strategy, its when the model cant count on the data being there in 3 days.

[Michael Tierney @Michael_WCD]

@alohaproxy A factory can only produce what it can measure and optimize for, so what metrics are you using to gauge its autonomous feature development and testing?

[Michael Tierney @Michael_WCD]

@ArunimaGanguly2 How many of those 100 images are just slightly different versions of each other? That's where most projects get stuck, not in designing them.

[Michael Tierney @Michael_WCD]

@arthuryuzbashew Those tools will only get used if they're easily discoverable, which means proper metadata on the internal registry. How are you exposing those tools to your org?

[Michael Tierney @Michael_WCD]

@CodeWithAmann Weve worked with a few clients who swear by Rasa, but for most small businesses the real challenge is getting their existing systems to talk in the first place, not choosing an AI tool.

[Michael Tierney @Michael_WCD]

@_naznotes Thats not our jam, were here for wiring together tools that dont talk, not spiritual strategies. We deal with CSVs breaking and APIs hanging up, not thanking forward or praising God in advance.

[Michael Tierney @Michael_WCD]

@Certified_Devvv Your launch page is already doing more to break the model than it needs to, 5 redundant Verifieds are enough for a human but will confuse an agent.