​

1. Agents are really just about clever prompting.

New concepts are shared every other week: workflow orchestration, tools, agent frameworks, autonomous agents, MCP, A2A, computer usage, and multi-agent systems. There’s so much noise it’s hard to make sense of it all.

Here’s what matters: at the heart of everything AI is an LLM, and that’s where the magic lies. Everything else, such as RAG, orchestration, MCP, and frameworks, is just plumbing. The entire industry is obsessed with this plumbing, creating clever wrappers around LLMs, but every framework interacts with LLMs in the same way we all do: in human-readable language.

This is profound when you internalize it. An agent is just a series of prompts that are sent to the LLM at different points to control personality, behavior, actions, and which systems to call. That’s it.

Whether it’s an agent taking over your computer screen or orchestrating a complex workflow, it’s all just prompts.

1. Speed: Code is slow to change. Configuration can be changed on the fly, and speed is everything.

2. Experimentation: Agent configuration needs constant testing and adaptation to ensure performance. Code is complex to experiment with.

3. LLM evolution: Models are constantly evolving, and agent systems need to keep pace. Agents in configuration can be changed in minutes, rather than hours, in code.

4. Prompt adaptation: Different prompts work differently with different LLMs. A prompt that worked today might behave differently tomorrow—you need to adapt quickly.

5. Developer Experience: Current agent frameworks are primarily for backend developers. This means that most developers never work on anything related to ‘AI’. Developers should have the opportunity to work with AI by doing what they do best, creating outstanding user experiences and integrating agentic experiences into their applications. To achieve this, they require lots of agentic workflows to integrate into the experience.

Winners are companies that enable their entire team to move fast with minimal resistance. This does not mean you should never use an agent framework either. If you are building a blockbuster feature, use them, create custom UI, and go all out. However, most of the features you need to ship do not require this.

To emphasise: being pro-configuration when it comes to agents doesn’t mean we’re a no-code system. If you read this entire letter, you will understand why we hold this view.

No-code tools are excellent for internal automations, but they are terrible for software companies building solutions in multi-tenant production environments.

Mindset keeps configuration where it belongs and code where it belongs. We’re an SDK product that provides customers with unlimited deployment controls.

​

2. Focusing on user experience and achieving outcomes is paramount.

Agentic applications comprise thousands of agentic workflows, including headless workflows that function like APIs, taking actions throughout your product, and user-facing agents that engage with people via chat or other UI components. That’s it.

Once you grasp this, everything changes. The differentiation isn’t what set of infrastructure and APIs you are using. It’s how fast you move and how well these workflows integrate into your user experience. The speed of delivery, the speed at which users achieve critical outcomes, and the speed at which they reach that ‘Aha’ moment is what counts.

Think of it this way: you have 10 fantastic ideas this quarter, but only build 1-2. Those ideas then need iteration because no feature is perfect on day 1. But here’s the thing- speed creates a compounding advantage. The faster you deliver, the faster you iterate, the faster you capture value, and the quicker you win market share in arguably the biggest land grab in history.

So, what if you could deliver 7-8 new excellent agentic experiences this quarter instead of 1-2?

That’s the power of agentic systems. They allow you to build and iterate at a pace that was previously unimaginable.

​

3. Subject matter experts should build agentic experiences, not just developers.

PMs used to write PRDs and hold endless alignment meetings while developers turned that noise into something that works. However, the Loveable, Replit, and Cursor revolution has begun to change this dynamic. PMs can now build prototypes faster than they used to write PRDs, and developers can ship experiences at unprecedented speed.

Yet most AI development tools still force an impossible choice: no-code toys or agent frameworks for mostly backend engineering only. A PM who understands customer personas and problems should be able to create agentic experiences, then work directly with engineers to deploy them in creative ways.

Think about Stripe—you configure it, then integrate it with unlimited flexibility. You don’t rebuild payment processing from scratch. Just like Loveable lets anyone create web pages and Cursor lets developers code faster, you should be able to define an agent’s behaviour, connect it to your data, build agentic workflows, and deploy it without writing Python.

This is what Mindset lets your SMEs do and is why we believe agents and agentic workflows are better configured than coded in most instances.

Companies that can iterate rapidly will discover what works while their competitors debate PRDs. Speed of deployment and value delivery are what matter most.

​

4. Post launch AI reality. There is no set-and-forget.

Here’s what happens after an initial AI launch (like your first date as a teenager): everybody gets incredibly excited about using new agent frameworks, RAG, or the other 3rd-party services you selected. You launch the feature, and then you go, “Okay, now what?”. We’ve seen many of those launches for AI, and we can tell you it means:

1. Self-service, building and customzing becomes everything

Like most B2B SaaS applications, enterprise customers don’t just want your feature- they want to customise it too. The same goes for AI and agents. Your customers will want to customise them and then build them inside your apps, using your data and existing features. They never just want the feature. They want to make things themselves, change something, alter something. Building a single AI agent and selling it gives you none of that flexibility. Salesforce got it right with Agentforce, which launched an agent builder with their agents - they knew that most users want to build or customize their own agents.

This need to customise constantly will never end, which becomes very tough when you have legacy features and other functionality you need to work on. It’s literally years of work.

2. Agents require constant optimization for outcomes and fast Aha moments

Like any new feature, they need refinement to meet user expectations and business objectives. The key is optimizing quickly: build, measure, learn, improve.

The challenge is that when everything is in code, optimization cycles become extremely slow. Your subject matter experts, who understand the business outcomes you’re trying to achieve, can’t make the rapid adjustments needed to improve agent performance. Instead, they’re dependent on development cycles to test new approaches.

The speed at which you optimize for outcomes determines success. AI systems need continuous refinement by the people who best understand what success looks like for your users. The goal isn’t set-and-forget automation- it’s creating systems that consistently deliver the outcomes your customers need. Real-time performance data becomes essential for understanding whether your agents are making a meaningful impact on the metrics that matter most to your business.

​

5. The canaries in the coal mines.

While most companies are still figuring out how to integrate AI features, a new generation of startups is allowing AI agents to control their core operations entirely. These companies aren’t just AI-powered—they’re 100% controlled by agents, and they’re growing faster than anything we’ve seen before.

They are the canaries in the coal mines: Gamma AI, Cursor, and Loveable.

These companies are not just building AI features; they are building AI-native products. They understand that the future of software is not about adding AI to existing systems but about creating systems that are fundamentally designed around AI.

We invite you to join us on this journey. The future belongs to those who move fast, empower their teams, and build agentic experiences that truly deliver value. Let’s shape the next era of software together.