Tyrone Robb

Marketing strategy, AI workflows, and better business decisions.

[Tyrone Robb @ty_auldric]

@MrSimonBennett I thought it was thunder and lightning today. I'll need to check my garden. Thanks USA

[Tyrone Robb @ty_auldric]

If this direction holds, the everyday dev probably gets both more help and more responsibility.

The good: security review becomes cheaper, more frequent, and closer to the PR. Smaller teams can catch issues they would never have had specialist time for.

The bad: attackers get the same leverage, and “AI checked it” can become false confidence.

The bar moves from “did you run a tool?” to “do you have the judgement, tests, and review loop to trust what it found?”

[Tyrone Robb @ty_auldric]

This feels like ABM at startup speed.

The kind of bespoke account work that used to make sense mainly for strategic enterprise accounts is becoming possible much earlier and at lower ACVs.

The artifact sells, qualifies, teaches positioning, and feeds product improvement.

What used to be first-100-customers manual work can now become a repeatable GTM system.

Small editorial note: I’d trim the blank space at the bottom

[Tyrone Robb @ty_auldric]

@JacobCEdmunds Either a poor model, or blanket. The full stop is an interesting artefact.

[Tyrone Robb @ty_auldric]

“Lean” is doing a lot of work here.

Some of this is probably payroll reduction. Some may be office footprint. But some is also substitution: founders using Claude Code, Codex, Cursor, etc. instead of hiring earlier.

That does not always show up as scary API spend either. A lot of exploratory building can happen on subscriptions before teams move to metered APIs.

The real question is revenue per employee, not just employee count.

[Tyrone Robb @ty_auldric]

I think this depends what kind of entrepreneurship we mean.

VC-style entrepreneurship can be a luxury: time, runway, safety net, risk tolerance.

But in some of the poorest countries in the world, self-employment is closer to the default because formal jobs are scarce. It is not even framed as “entrepreneurship”.

The real distinction is opportunity entrepreneurship vs necessity entrepreneurship.

[Tyrone Robb @ty_auldric]

@bryantchou It was terrible for me, prelaunch, no website. Would be nice to have a provision for day 0.

[Tyrone Robb @ty_auldric]

@GeorgeJeffersn I wonder how many YC founders from former batches think it’s a bad deal.

Calling YC expensive from the outside is a bit like saying the steak looks overpriced when you’ve only ever had bacon.

[Tyrone Robb @ty_auldric]

The business issue is not whether inequality is real.

It is what kind of tax design changes behaviour.

If the policy hits illiquid paper gains, you risk pushing founders, investors and capital allocators into defensive behaviour.

Better question: how do you raise revenue without making productive risk-taking less attractive?

[Tyrone Robb @ty_auldric]

@ShubhamInTech @asantanillaj @GaryGao891219 @ianyfong Was it the hard work at the start up or the gym that made you crash?

[Tyrone Robb @ty_auldric]

I’d separate the signals.

A complaint means interest.
A workaround means pain.
Existing spend means budget.
Repeated urgency means timing.
A bad paid alternative means an opening.

“People will pay” usually starts when at least 3 of those show up together.

For your X growth/AMA angle, the signal probably isn’t “people want followers.” Everyone says that.

The stronger signal is whether they’re already buying courses/tools, spending hours replying, failing to turn attention into leads, or asking the same tactical questions repeatedly.

That’s where advice starts becoming a product.

[Tyrone Robb @ty_auldric]

This is especially true outside tech.

I spoke to an SMB owner recently who talked about AI like they were using full ChatGPT/Claude, but their actual workflow was free Perplexity and Bing.

A lot of people are judging the category before they’ve used the thing the category is becoming.

[Tyrone Robb @ty_auldric]

This is especially true outside tech.

I spoke to an SMB owner recently who talked about AI like they were using full ChatGPT/Claude, but their actual workflow was free Perplexity and Bing.

A lot of people are judging the category before they’ve used the thing the category is becoming.

[Tyrone Robb @ty_auldric]

@CreeCoder Is the goal monetisation, or are you trying to funnel attention somewhere else?

Verified impressions matter for payouts, but if the goal is customers or leads, profile visits, replies, follows, and conversions probably tell the better story.

[Tyrone Robb @ty_auldric]

@pcshipp replying to you and others on X.

[Tyrone Robb @ty_auldric]

@pmitu When everyones tech is tool old to run agents and the cost of new tech plus their AI sub is more than their annual rent.

[Tyrone Robb @ty_auldric]

The pricing pressure is real, but the answer is not simply “use the cheapest model.”

The winning tier is probably robust-enough intelligence priced low enough that teams don’t ration usage. Frontier for judgement, cheaper models for repeatable work, and a harness that routes between them.

Also, some teams may need to accept a bit more workflow jank before jumping straight to raw API usage. Codex/Claude subscriptions can be far cheaper for exploratory work. I’ve seen an intern burn $2k on AI API usage.

The issue is not just model cost. It’s the commercial model around ai coding.

[Tyrone Robb @ty_auldric]

This pattern shows up whenever the hard asset is not the product, but accumulated trust and context.

Unilever buying Dollar Shave Club was not because it couldn’t make razors. PepsiCo buying SodaStream was not because it couldn’t make drinks. L’Oreal buying CeraVe/NYX/now Innovist is the same logic: buy the behaviour, community, channel knowledge, and credibility already formed.

The opposite is also instructive. Big companies often fail when they try to build challenger behaviour internally. They can copy the product, but not the cultural permission, trust, or speed that made the challenger work.

[Tyrone Robb @ty_auldric]

The answer is probably not Claude Code vs Codex.

It’s the harness. Something like T3 Code/OpenCode style routing makes more sense if you want to swap models by task: frontier model for judgement-heavy work, cheaper models for repeatable/checkable work.

The dog is right to ask for compute, but should not get merge permissions.

[Tyrone Robb @ty_auldric]

Shipping is the fastest way to learn, but only if each project is testing one distribution assumption clearly.

Otherwise you learn “this worked” without knowing whether it was the channel, offer, audience, pricing, timing, or packaging.

Directories can be useful, but the signal is noisy unless you track what actually converts after the click.