Gif by Ruposh on Giphy
I changed up the format a little bit this week and have a shorter newsletter with some quick practical application, several links to AI things I’ve been reading, and a quick prompt to improve your output with AI. Let me know what you think 😊
Skip the DVD
Last week in a client meeting, somebody said "Hey, it'd be cool if we had this." Five minutes later it was live. Not a feature request, not a backlog item. Live.. while we were still in the meeting.
This is the new baseline.
When I started working in the tech/software space years ago, the answer to "we need a tool/app" was to start a months-long journey of discovery/research + sessions to gather/define/prioritize requirements + lots of designing and prototyping + user testing + software development + QA.. and THEN (finally) launching the app. Then you slogged through fixing bugs and tackling feature requests and integration work and maybe 3-5 years later, you replaced/rebuilt it.
That cycle wasn't a choice, it was the cost of the materials. Software was a physical thing you bought, like a VHS tape or a DVD. You spent a lot to make it, you protected it once you had it, and you kept it on the shelf until it broke.
That has collapsed, just like how the VHS and DVD market collapsed when streaming came out, except we skipped straight past the DVD part..
I'm not an engineer but I built the entire StealthX Armory, which is our business operating system for the company. It has dozens of capabilities including our full CRM, complete project management, staffing management, and a whole slew of other tools, agents, and data we need to run the business.
The implication for your business or team is simple.. if you're still planning with long discovery, locked requirements, months of development, and a 3-year shelf life.. you're solving last decade's problem. The AI era doesn't just mean faster. It means the output itself is completely different. It's not durable, it's not precious, it's a session with AI tools, not a product.
What it actually takes to do this
Here are some specific ways to put this into action within your company/team/org:
Pick the constraint.The shift from 12-month projects to shorter 6-week sprints isn't a scheduling change, it's a planning change. Stop scoping the work to a fixed feature list with a flexible date. Scope it to a fixed date with a flexible feature list. Ask "what's the most we can ship in six weeks that creates a visible outcome?" and let your team negotiate the rest. The constraint forces clarity faster than any deck or document ever did.
Use buckets, not Gantt charts. Under real ambiguity (which is what your AI goals are), trying to plan every dependency in detail is pretty comical. Focus on buckets of work and iterate/plan weekly, not monthly. Make ambiguity easier to navigate by using buckets instead of pretending it's clear.
Hire for tempo, not just skill. Your AI initiatives don't need more headcount. They need people who can move at the new tempo. If your last hires were great at the old cadence, you've already over-hired for the world that's leaving.
Things I'm reading
Lenny Rachitsky's takeaways from Claude Code's Head of Product. Five-minute read on how the Claude Code team actually does product. Worth your time if you're anywhere near a product org right now.
1,116 Claude Code skills, catalogued across 500+ repos. Someone did the legwork so you don't have to. If you're building with Claude Code, this is your library card.
A task delegation prompt block for sub-agents. If you've been wondering how to actually delegate to sub-agents instead of one giant prompt, this structure is the one I'm running with.
The CLAUDE.md file that 10x'd one builder's output. A configuration template that's better than what most folks (including me, six months ago) start with. Steal liberally.
Portless: replace port numbers with stable named URLs. Tiny tool that swaps localhost:3000 for a name you can remember. Stops me from forgetting which port is which when I have four projects running.
The Claude Code team responds to recent quality concerns. Honest, technical, useful for context if you've felt the slip in your own workflow over the past few weeks.
design.md: a visual-identity spec for coding agents. A format for telling AI how your visual identity works. Could become a standard for the designer-to-agent handoff.
Weekly AI tip
AI tools generate plans faster than humans can read them, but just because it’s a plan doesn’t mean it’s the right plan. Before I let any agent or AI tool execute anything, I have it plan and I also have it generate 3 to 5 verification questions about the plan and answer them independently. This week alone that habit caught 5 errors in an agent's report (wrong file paths, wrong counts, a "dead code" claim about a file that was actually live) before any code shipped. Takes a few extra minutes but saves hours of troubleshooting.
Here’s the actual prompt I use get better outputs:
[Insert your goal/request]. Analyze and create a plan to address for me to review/approve. Then generate 3-5 verification questions that would expose errors in your plan. Answer each verification question independently. Then provide your final revised answer based on the verification.
Onward & upward 🤘
Drew