People don't want your app, they want results.

I just dropped a recent fireside chat with my thoughts on the landscape ahead after some recent announcements in the AI universe. If you prefer to listen, check out the episode below.

You can also listen to the episode on Spotify or Apple Podcasts. Here are some of the key takeaways.

The shift from apps to agents.

There’s an incoming shift from people using your app or website to solve their problems, to solving their problems in the AI tools they’re already using every day. 2 weeks ago at OpenAI’s Dev Day they announced their Apps SDK and Agent Builder.

If you missed it, basically you can now hook your app or software into GPT and give your customers the ability to take action directly in GPT. They shared a few demos of apps like Spotify and Coursera during the event that signal an incoming paradigm shift..

Soon people will stop thinking about features they need from your app or information they need from your website, and they’ll start to think in how their AI platform of choice can do that for them without ever needing to use your app/website.

What this means is that every company will need to start shifting their digital experiences to a conversation in another platform. People are going to start asking themselves “Why do I have to use this app? Why can’t I just do it here in GPT (or Claude or Gemini)?”

The work for product teams now is about figuring out how to give customers/users access to their app’s capabilities within a different modality. If you’re a SaaS company, you should be very quickly trying to build MCP servers for your app/software to connect to these platforms.

Not sure where to start? I’d suggest defining the top three things people are trying to accomplish in your app/software/website (i.e., the customer’s “jobs-to-be-done”) and start working through how you can enable folks to accomplish this using your data and capabilities directly in GPT (while still driving revenue for your company).

Which leads me to the next topic…

Everyone needs to spend time thinking through monetization in an AI world.

If you’re allowing your app/data/capabilities to run inside someone else’s platform, how will you make money?

Treat AI & agents like their own sales channel with real unit economics. Track latency and success rates. Consider pricing for outcomes or layering in usage or capacity, not just seats.

Think about the new upgrade paths that make sense to customers. This might be faster models, priority execution, higher action quotas.

Good news is that everyone’s trying to figure out the new models for monetization, so you’re not behind (yet). I strongly encourage you to spend some time thinking about this so you’re not caught flat footed a year from now.

People’s willingness to change is the real blocker (it always is).

Right now, teams aren’t blocked by technology. They’re blocked by each other. They’re blocked by broken and undocumented processes, fragmented data, misaligned mental and incentive models, lack of trust and crappy culture, and fear (of change, failure, looking stupid). I am seeing this with every company I talk to right now.

People don’t want to change.

There’s a term going around right now called “work slop.” It’s becoming a common counter argument for why AI isn’t good enough to use or trust in the work environment.

Fundamentally I disagree.

If you’re unfamiliar with the term, work slop is the phenomena where one person uses AI to do their work without critical thinking or evaluation, then pass it on to someone else to deal with. Work slop happens when someone creates a quick output with an AI tool and don’t spend any time or energy to assess if it’s actually good/correct/useful. They pass it on, assume it’s good, and expect the next person will figure it out.

I acknowledge that work slop exists (I’ve seen a lot of it), BUT it’s not because AI isn’t capable of producing good quality work. It’s because people haven’t been trained on how to think and use AI effectively.

Fundamentally AI requires a fundamental mindset and behavior shift across an organization that’s solvable with the right approach, training, and coaching.

To drive this change, teams need to learn how to:

• Ask AI better questions and how to word them in a way that gets the best results.

• Break down complex requests into step-by-step actions.

• Use the right AI tool to get the best results for different goals/needs.

• Share the right level of context (the Goldilocks principle applies here.. not too much, not too little)

• Think in systems and act like an architect and orchestrator. Don’t just dump a question in and expect it to infer or interpret what you mean.

Collaboration in AI tools has gotten WAY better.

Last year it was a HUGE pain to try and collaborate in most AI tools. For example, in ChatGPT you could create a project that had a set of your chat threads and sources, BUT you couldn’t share that project with anyone else on your team. Huge missed opportunity. Fundamentally, teams need to work in the same space so context is shared and compounds.

Thankfully, over the last several months team collaboration in AI tools has improved dramatically (although it’s still far from where it needs to be). Now you can collect and share every note, transcript, thread, prompt, and source in one project for everyone on the team. You can do this in all the major model providers (e.g., GPT, Claude, Gemini), which has made working with a team in these tools massively more efficient.

Another great example of a great AI collaboration tool is Google’s NotebookLM. If you haven’t tried it out yet, USE THIS NOW. Once you’ve created a “Notebook” (like a project in GPT), you can share with others and collaborate in the same workspace.

You can quickly and easily upload or link different docs or content as a source that the model uses as its knowledge base. You can chat with it like all the other AI tools, BUT it can also create outputs in other mediums like a podcast, video, report, or quiz. Such a helpful tool when trying to quickly ramp someone else up on a project or remind yourself of past conversations.

We’ve been using for client projects and it’s been a massive time saver to collect all meeting transcripts, notes, and artifacts in one place so anyone can “talk to the project” and use as a project copilot.

One other nuanced thing with NotebookLM is that you can link live Google Docs and sync periodically vs manually deleting outdated docs (aka “sources”) and uploading new copies. This is a little bit better than GPT and the other big model provider tools, but none of them have released realtime file/folder syncing in shared team spaces (yet). Can’t wait for this to be added 🤞 

We’re testing and using all of these collaboration features daily at StealthX and they’ve been a massive time saver.

Start with tiny agents that remove daily friction.

I’ve been facilitating a lot of AI workshops with teams that want to jump straight to tackling complex workflows with custom AI agents.

Don’t do this.

Start very small. Build a tiny agent that simply converts meeting transcripts and notes into follow up emails in your voice. Then pick the next repetitive task. Rinse and repeat.

Tiny agents are incredibly valuable and help build confidence and credibility within a team or organization. People can feel the savings immediately and leaders can easily measure the impact before/after.

Start with quick wins and incrementally move to custom/complex agent-powered workflows.

Efficiency is table stakes. Trust & differentiated experience will win (especially for the mid-market).

Very very soon, basic AI tool usage will be status quo (arguably we’re already here).

The window of time to use AI tools as a competitive advantage is shrinking by the day and in within the next 6-12 months it’ll be the new floor to keep up with the market.

At StealthX, we believe the edge for all companies will become a differentiated brand and experience that is unique, memorable, and trustworthy. This is true for all companies and all industries. Ultimately, you’re still serving humans on the other side of the screen 😉 

This matters most for mid market companies.

Startups can move fast and pivot daily. Large enterprises can sit tight and ride out the AI upheaval with their existing customer base, brand equity, and perceived switching costs for their customers.

The middle is going to get squeezed even harder UNLESS they can create a memorable experience, build strong trust with their customers, and clearly demonstrate what makes them different/unique. If you’re a mid-sized company, invest now in human touches while making things more efficient with AI in the background.

Wrapping up

As usual, I love to make this newsletter super practical so you can apply the lessons in your company and in your workflow, today.

1. THIS IS CRITICAL: If you haven’t already, CLEAN UP your data. Then start working on setting up MCP servers for your top 3 customer actions/activities/goals/needs.

2. Teach your team how to think and use AI. Train them to break work into steps. Teach them how to ask better questions. Give them simple templates and show real examples from your own work. Set clear expectations for everyone to do quality checks along the way to avoid work slop.

3. Start small and build 1 agent. Pick 1 team and 1 repetitive task. Build an agent that can handle this task for them, then measure time savings. Share the before and after with other teams to build awareness/excitement, then build the next tiny agent.

4. Map the 3 highest trust moments along your customer journey. Ask yourself, “Where is the biggest opportunity to connect with our customer?” Replace generic touches with specific, personal gestures. Think and act like a real human, NOT a faceless organization by partnering with creators in your niche, investing in communities adjacent to your business, and showing real work with real people from your company in public.

Onward & upward,
Drew