Henry Mao

Co-founder/CEO @ @SmitheryDotAI, backed by @southpkcommons;
Prev. co-founded (exited)

[Henry Mao @Calclavia]

The future of the internet will be dominated by tool calls, not clicks.

We're building Smithery to orchestrate this new era of AI-native services for AI agents.

Read our mission 👇

[Henry Mao @Calclavia]

@abhiaiyer I find that having some structure to BDD helps.

Gherkin is one such structure that is easy to read and from user POV, but maybe there's a new syntax we've yet to discover

[Henry Mao @Calclavia]

Gherkin is now making a comeback.

If you haven't heard of Gherkin syntax, it's a classic way of writing behavior tests in plain language:

Given [I have an account on X]
When [I publish a post]
Then [it should appear on the timeline]

I learned about Gherkin briefly in an agile development class back in university. I tried this approach at my previous startup, where the PM would write Gherkins and hand them off to engineers. That way, the PM could focus on the user journey and desired outcome without needing to be concerned about technical details.

This process and TDD/BDD never really stuck in practice due to overhead maintenance and brittleness, but things are now changing.

With agents being so good, we've now moved towards defining /goal based optimizations and no longer write code. To do that effectively, the system needs backpressure, and humans have to be on the edge, rather than in the loop.

It's becoming critical to make systems verifiable with tests and requirements. Gherkins just happen to be a neat, structured way to define all of these tests that's human-readable.

We're now bringing back Gherkins and getting models to walk the test cases. I'd recommend revisiting some of these old ideas!

[Henry Mao @Calclavia]

@abhiaiyer Doesn't literally have to be Gherkin, but I find outcome/behavior driven development increasingly the right place to work.

What's your take?

[Henry Mao @Calclavia]

The Mercury MCP is now on Smithery. @mercury

We've been using it to check our accounting and find wasted expenses.

Try it!

https://t.co/XuBq3E0i0i

[Henry Mao @Calclavia]

@RhysSullivan How does it compare with https://t.co/taaQ2uB5xD?

[Henry Mao @Calclavia]

Prompting is the tax we pay for agents that lack agency

[Addy Osmani @addyosmani]

https://t.co/0DyFIXJueI

[Henry Mao @Calclavia]

Maximizing token spend works better if we're also minimizing the time spent prompting, steering, and reviewing agents.

[Henry Mao @Calclavia]

@thsottiaux Should bring GPT 5.5 Pro!

[Henry Mao @Calclavia]

@grinich @Cloudflare @firecrawl @WorkOS Amazing to see more work in this area. It's a huge pain point for Smithery users to have to sign up to obtain API keys, and we've been experimenting with ID-JAG.

It's great to see this being pushed forward!

[Henry Mao @Calclavia]

Brandfetch MCP is now on Smithery! @brandfetch

I've been using it to find high-quality brand icons/logos to display on decks and our pages.

Check it out:
https://t.co/Gv7gpK06xJ

[Henry Mao @Calclavia]

Thanks to @buildclub_ @AnnieLiao_2000 for having me talk about how we can move beyond the world of manual prompting towards agents with more autonomy.

Was great to demo with @SerenaLamAU, @picocreator, and @fatimarizwan!

[Annie ❤️‍🔥 @AnnieLiao_2000]

excited to share that @buildclub_ community is officially live in Singapore in partnership with @singtelinnov8! 🇸🇬

last night, 250 builders gathered to witness the launch

thanks Serena Lam, @picocreator, @Calclavia and more for demoing!

if you want to get involved, we have some incredible exciting things in store (please reply or reach out!)

[Exa @ExaAILabs]

We raised $250M in Series C funding at a $2.2B valuation, led by a16z.

Exa is a search lab organizing the web's data for agents.

[Henry Mao @Calclavia]

@ExaAILabs Congrats!!

[Henry Mao @Calclavia]

Here's how I got coding agents to relieve my allergy:

1. Connect it to all my contexts (via Smithery)
2. Spin up a long-running task via /goal that runs for 30 min+
3. Be bored
4. Realize I need to do higher leverage tasks: something agents can't do
5. Start cleaning my room, removing a substantial amount of dust
6. Checks agent => it's still running
7. Decides to clean the bathroom too
8. PR is shipped. Allergy is gone.

Insane increase in shareholder value.

[Henry Mao @Calclavia]

"It's our strongest model yet"

[Henry Mao @Calclavia]

Another analogy I heard from @GeoffreyHuntley at @aiDotEngineer is that vibe coding is like giving everyone an iPhone camera.

It didn't put wedding photographers out of business.

[anirudh @kamathematic]

vibe coding is cooking at home; software engineering is being a professional chef. access to kitchens didn't put restaurants out of business

[Henry Mao @Calclavia]

Thanks to @SherryYanJiang, @agrimsingh, and the organizers/speakers/sponsors for making @aiDotEngineer Singapore happen over this weekend, letting me go into the rabbit hole of MCPs/CLIs, and how to wire up your agent.

Met a lot of new people and reconnected with many friends!

[Henry Mao @Calclavia]

We're settling the MCP vs. CLI debate by benchmarking Codex and Claude Code across 3 different APIs, totaling 756 runs.

We also cover skills, code mode, pretraining bias, and how MCP fits in the broader picture.

The results might surprise you👇

[Henry Mao @Calclavia]

@jxnlco UI still makes me go to Claude

[Henry Mao @Calclavia]

I'm back in Singapore and will be speaking at @aiDotEngineer this Sunday.

DM me if you want to hang out at the conference this weekend.

[Henry Mao @Calclavia]

If models and harnesses get even better, it's plausible to imagine labs will stop selling tokens as APIs and just sell outcomes.