Ask The W

Every fast AI-native team has a graveyard of decisions nobody can explain anymore. A tiny whodunit about what happens when the why goes missing ▶️ Psst… the detective is askthew.com

@darrenmarble Thankful to have solo founder friends who help out here!

Technically, you don’t need a cofounder now. Emotionally, it might be smart to have one.

@willchen500 Contrary to popular belief, the harness/UX around the model can be a moat. Google flights existed but so did Kayak and Expedia. There's a dozen CRM tools that have done well. Users have personalized taste and the companies that will focus on CSAT, will win a piece of the pie.

Harvey and Legora are probably thinking of pivoting to law firms. Or at least acquiring an “AI native” law firm subsidiary like Carta. It’s the only way that they can provide a narrative for how they will survive, now that: 1) Biglaw firms are building their own AI application layer/fine tuning models; and 2) OpenAI and Anthropic are going after their customers and they don’t have a differentiated product. But the moment they do so, they will enrage all of their customers who all have opt out clauses embedded in their contracts, thereby triggering an immediate ARR collapse. Truly a difficult dilemma.

As a former PM who loved talking to users, I've never been more excited about building. At an event this week, I fixed 4 product issues directly from my phone while talking to users. For years, PMs optimized for alignment: research, buy-in, PRDs, stakeholder reviews, roadmaps and endless coordination. Now we're moving into an era of build, ship, talk to users, iterate and get paid. Builders get to be closer to the customer than ever before so it's a great time to be a PM (or whatever the new title is!)

I personally consider it very stupid to say that the PM role is going away. That is FUD with zero substance. Anyone who wants to be smart right now will recognize something obvious: if you have a product idea, you can just build it. Anyone who wants to continue being smart right now will also acknowledge the multitude of downsides and risks that come with this empowerment. They are obvious. Especially if you just build something. You’ll see it. If you are in product, just use AI and pay attention to how it goes when you do. If you do, you’ll automatically be in the future state, armed with skills, opinions, and wisdom. It doesn’t matter what is going to “happen to the role.” If it’ll have a new name, a new scope, go away, or whatever. What matters is if you are personally competent or personally incompetent at the future state. Ignore the FUD. Engage with the substance.

This is one of the 'aha' moments from a prospect who heard about Ask The W! Decisions previously made due to political pressure can be eliminated by connecting agent decisions to outcomes.

@vasuman Just posted about it earlier. Opus 4.8 works better on Claude Code imo. x.com/AskTheWHQ/stat…

Something went very wrong in making Opus 4.8. Talks exactly like 4o. Hallucinates very hard.

Used SuperGrok, Grok Build and @premium X this week and seriously this bundle is mind blowing 🤯 The only thing left for X is to build an Agentic browser for automation.

You're speaking our language! We turn signals from wherever work happens: Claude Code, Codex, GitHub, Slack, Linear, Browser and more. We surface key decisions, contradictions and next moves. Built for both technical and non-technical teams that are rapidly becoming AI-native. askthew.com

This is effectively the #1 problem for AI agents in the enterprise. As we go from agentic coding (where a large amount of context is in the code base, and users are technical enough to get the rest to the agent easily) to a world of knowledge work agents, the context problem becomes much more acute. We see this every day with customers at Box. For existing digital knowledge, it’s often fragmented across legacy systems or environments that don’t play nice with agents, and have access controls that don’t map to the real work that needs to be done, which become a huge hurdle for getting agents the context they need. This has to all get moved to modern, secure cloud environments. But also, companies often haven’t captured and digitized some of the critical context that agents need to work with. Decisions, processes, and workflows often live in people’s heads and tribal knowledge that need to get turned into unstructured data for agents. This is actually one of the biggest points of leverage for applied AI companies, because they can work to specialize in getting agents exactly the information and domain expertise they need. But it’s also one of the reasons why FDEs and new system integrator plays will also work so well right now. The companies that figure this out will be able to get the most out of AI going forward.

@Jayyanginspires Solo builder here! Count me in. Happy to share.

I feel like I need to host a weekly call where 3-5 of my friends all screen share for 30min and show how we're using AI in our work / businesses. I have so much AI guilt. Not using it enough, powerfully, or correctly.

@NotionHQ The judgement layer for an AI-native team askthew.com x.com/AskTheWHQ/stat…

What are you building today?

This is why you need askthew.com

@emollick That's why I'm building askthew.com 😊 Making sure humans and agents ask the right questions over time.

/goal and other fully automated AI agents are cool, but not a great model for the future of work with people. Instead you want your AI to know when to ask you GOOD questions, maybe because it is stuck, maybe because your taste matters, maybe because you would find it interesting.

@aryanlabde Spending Sunday helping buyers understand the value of the product. x.com/AskTheWHQ/stat…

What are you guys working on this Sunday? Pitch your product. Yes count this as marketing.

@eniac We separate signal from slop so that the bullet points you receive are not AI slop as well. askthew.com helps teams stay focused while building with AI tools.

Somewhat unpopular opinion - AI is becoming a HUGE distraction. Giving everyone the power to build, without first forcing clarity on what actually needs to be built, creates lack of focus and an abundance of AI slop. That’s why I’ve started asking people to send me bullet points instead of polished AI-generated presentations. Bullets are easier to digest. And ultimately, if I need the full deck, I can ask AI to build it myself. In an AI-first world, focus matters more than ever.

@harjtaggar This is the way to go. I built 46 prototypes on Free/$20 plans before one eventually made money. Beyond revenue, the process gave me the confidence and conviction to fully commit to my current startup.

Finding a startup idea is usually a mix of breadth-first and depth-first search. You look at a bunch of branches, then force yourself to go deep on the one that seems most promising. Repeat until something pulls you in. Tokenmaxxing helps with the breadth-first part. But the hard part is still building enough conviction to go depth-first for a while.

@sflorimm Making good decisions

Which skill do you think will matter most in the next few years?

💯 At askthew.com, I think about the compounding effects of decisions humans and agents make over time. Dev making critical decisions on a monday, forget on friday and teammate (+ agents) lose that context in 2 weeks. This system is not something the dev should worry about!

Compound engineering is already a term people are using, and the definition still feels a little emergent. But one specific version of it that I’ve found useful is this: Use agents, automations, callbacks, and scheduled jobs to keep adding small bits of utility to your engineering process without needing to remember to do them manually. A practical example: For some codebases I’m working in, I have a weekly GitHub Action that runs Claude Code in non-interactive mode with -p. It uses a few custom skills that know how to inspect the repos, look at recent changes, and understand the broader system. From that, it generates a few markdown files: - an architecture overview - a recent code changes summary - a higher-level view of how different repos or services fit together Those markdown files get deployed into a repo where I also have a Claude Code plugin, so future agents can access that context directly when they’re working in the system. You could ask an agent to figure this out dynamically whenever you need it. But I’ve found it useful to distill this context ahead of time, so future agents have a cleaner map of the territory. I’ve also added another step that takes those markdown files and turns them into a self-contained HTML/CSS/JavaScript site. Once it’s deployed, you can click around and actually explore the architecture interactively. It becomes a condensed way to read the distilled architecture of the system instead of digging through a long set of markdown files or trying to reconstruct it from the codebase. This has been more useful than I expected. It’s helpful when I’m trying to understand how a specific part of the system fits together, and it’s also useful to share with people across the team who are working on different pieces of the codebase. It takes some tuning. You have to build the skills, apply your own taste, decide what the architecture docs should include, and decide what the HTML output should look like. But with an agent helping, it’s not that much work. And once it’s wired into GitHub Actions, it just runs. I think this is the kind of thing people should be looking for when they build with AI. Not just: “what can I get the agent to do right now?” But also: “what small system can I build once that keeps making this process better without me thinking about it?” Not everything needs to be a chat interaction. Some of the most useful AI work I’ve done lately has been building small background systems that improve the context, documentation, and workflows around the thing I’m already building.

@boardyai Flexed my creativity with Grok Build for an upcoming demo. x.com/AskTheWHQ/stat…

What did you build this weekend?

When you export a transcript, what are the most important things you want to preserve? And how do you currently find and surface those decisions later? Overall, I agree with your sentiment. I put together a lightweight plugin for Codex and Claude Code that securely captures key signals as transcripts are generated. These signals + GitHub commits/PRs + other input sources curates decisions, next steps, contradictions and the surrounding context later. If this is important, I’d love for you to try it out askthew.com

I'm really upset about this: OpenAI's Codex Desktop had a "Copy as Markdown" option for exporting full chat transcripts, but the feature vanished in an update a couple of days ago Genuinely my single favorite feature of Codex compared to Claude Code github.com/openai/codex/i…