IIIMPACT Design

"A designer is not an artist. A designer is a commercial problem-solver, hired to take a client’s half-formed intent: a vibe, a feeling, a vague sense of “make it look professional”, and give it form.

That is the job whether the tool is a pencil, Photoshop, or a prompt box. AI has not changed what a designer does. It has changed who can now attempt to do it themselves. "- Joshua Leigh

Everyone keeps talking about distribution as the moat for your products.

We know speed is table stakes now. Everyone has speed. Quality...mmm not so much.

With physical products that can go viral it definitely helps. With software products, with AI, you should be able to ask it to find apps to do relatively easy unless it's brand spanking new.

So then it comes back to UX.

Is the design enjoyable, frictionless, solves my problem easily for the value that it costs and I don't want to vibe-code, maintain or switch to a competitor.

So I ask again, is Distribution really the moat...or is it the UX?

We developed a UX / AI agent algorithm system so powerful it's too dangerous to share publicly. Let me know if you want in.

(Hey don't hate the playa...works for anthropic/openAI...maybe might work for us).

We're compiling code faster than trust.

AI amplifies software engineering experience. Seniors gain the most; juniors use more tokens with less results.

AI agents are the new jr. engineers/designers but they need guard rails.

Cap your daily agent output to what you can actually review and verify

Tech/Design debt is cheap to take care of with AI - so use it to clean up your product.

Right now it's too easy to let AI write the code or design the app while you skip the learning. The bug or problem gets fixed but your mental model doesn't move

Spend more time thinking and understanding as we are trading future capability for present-day speed, and these tools won't make us do otherwise.

From he age old saying - if you have 60 min to solve a problem, you should spend 50 min understanding the problem.

Slop buries the people who care.

- more, longer, pull requests
- most devs have code remove fatigue, so just say 'Looks good to me!' and don't remove it.
- just a small number of devs push back on AI slop
- devs who care are being overwhelmed
- some of these devs are getting fed up, burning out

CEO of Opencode - "We're shipping way more hacks (code quality is low)... where we should have just rethought the whole system from the ground up, or redesigned it to make it more flexible...so I think our judgement is just off."

"It sure feels like software has become a brittle mess, with 98% uptime becoming the norm instead of the exception...and user interfaces have the wierdest bugs that you'd think a QA team would catch. I give you that that's been the case for longer than agents exist. But we seem to be accelerating."

Don't know how many enterprise software companies I've spoken with over the past 2 decades that were just hacking crap code with people...now they are doing it with AI.

Doing this without the expertise to know how to fix or find anything in the code because they've laid off or their best talent has left, is going to sink their software ship faster.

An AWS exec said something recently that stuck with me:

"Organizations that are moving faster and maturing do something different. They begin with a business outcome and work backwards, building an agentic system that reduces decision handoffs, increases learning, and removes friction, all while maintaining quality. AI maturity comes from applying AI to the system*, not to the individual."

Meta was awarding devs that used as many tokens as possible, which they just recently removed this poor incentive structure.

Here's what we're seeing on the ground:

Organizations are generating more AI slop, more bugs, and more tech and design debt, faster than ever. Their QA teams can't keep up with the output, so now "the devs do QA." They've turned people loose with AI in the name of speed, with no governance underneath it.

That's the exact opposite of working backwards from an outcome. It's a recipe for disaster.

Our team has been working relentlessly with organizations to do the counterintuitive thing: slow down to speed up. Break down internal processes. Build the AI harnesses and governance that keep UX and product strategy from becoming an afterthought, or a one-shot prompt in your AI tool of choice.

"But I can one-shot a design that's close enough."

So can your competitors.

The edge isn't speed. It's discipline. It's building data-driven agents with feedback loops, pulling from customer service data, user behavior, analytics, that give your prompting-happy PMs a more rigorous path to a better product.

Start with the outcome. Build the system. Everything else is just slop at scale.

New Anthropic policy: "To ensure we're responsibly deploying Mythos-class models, we are requiring limited data retention and review as part of our safety work. Prompts submitted to, and outputs generated by, Mythos-class models are retained for 30 days for trust and safety purposes, on every platform where these models are offered. "

If you use fable/mythos - they collect your data ...no exceptions even for enterprise partners.

Even if your organization previously negotiated zero-data retention agreements, you are now subject to this new 30-day window.

Advanced power now comes with a new set of data governance trade-offs.

Anthropic is like an abusive parent who buys you the best toys so you just deal with it.

Claude Claude is Anthropic's AI, built for problem solvers. Tackle complex challenges, analyze data, write code, and think through your hardest work.

Stop looking at cost per API call. Measure the Cost Per Successful Outcome.

Moving faster in the wrong direction, is far more expensive than moving slowing in the right direction.

When Compute Costs More Than Headcount

We are officially seeing AI compute expenses actively outweigh the personnel costs they were supposed to replace.

The biggest mistake happening right now? Treating AI like a static SaaS license. AI is a variable expense. Poor usage governance recently cost one enterprise $500M in a single month because they didn't cap employee access...but hey you all have to move faster and use "more AI".

The token costs for running AI agents are now exceeding what they were paying the employees they fired.

When the tokens run out, the AI stops. Just stops. No continuity. No workaround. Just a spinning wheel where your workforce used to be.

You fired humans to save money and bought a subscription that bills you into a corner.

The employees you let go knew what to do when things broke.