Luca Grulla |

Agentic AI removed the escape route

2026-03-26T00:00:00+00:00

Building software has never been so easy, and yet leading technology has never been so hard.

The velocity of execution requires a review of the more fundamental principles of how the technology organisation operates.
The expectation from investors has skyrocketed.
And if you are not an AI-native business, you have pressure to understand how to compete in the “AI eats the world” phase.

It feels daunting. And things keep changing. Where to start from?

The principles you need were written down twenty years ago. Most teams ignored them because they could get away with it. Agents have removed that escape route.

For many teams and organisations, the pillars were:

Planning: you need to understand what to do and what the cost would be, as you can fit a limited number of experiments in any given time. Planning will enable throughput and consistent delivery.
Total Cost of Ownership: what’s the overall cost of this initiative? In particular, will this have a long tail of micro work that will eat away precious planning slots?
Quality: manual inspection via code reviews, testing, human judgment was how you kept standards high. At human-produced volumes, it was imperfect but just about sufficient.

If you could manage these three steps, you would be well placed to achieve your goals.

But the landscape has changed. When the cost of writing software is shrinking, and the time to execute shrinks from days to hours, what is the north star?

Ducati 98 assembly line, c.1957. Public domain via Wikimedia Commons.

Today, what matters is:

Clarity on the goal: as execution time shrinks, clarity on the goal is more important than ever. Run more experiments than ever before, but be ruthless about which ones graduate. Small steps are great for agents and for focus: smaller prompts, tighter scope, better results. Repeat.
Repeatable feedback loops for sustained velocity and sustained progress: your agentic coding flow is 10x better if you empower agents with a deterministic test. It’s teaching your agent what the final goal is, not letting the agent guess it. What was (wrongly) often seen as an unnecessary burden in the past is now a hard requirement, or the precious tokens will be gone.
Observability is what enables quality and sustains velocity. Code reviews have always been more of a confidence mechanism, not an effective quality raiser. It worked well enough when humans wrote code slowly enough to read it. We passed that stage. Furthermore, the bugs that actually hurt you in production were never visible in a diff. Everyone is now pushed towards observability, a mechanism that many teams should have trusted more all along. Finally, models are exceptional at identifying complex incident scenarios — making observability the connective tissue between production reality and agentic incident response.

The agentic revolution, by stripping away the implementation bottleneck, is a forcing function for how teams and businesses operate. Old operating models were built on assembly line logic — optimise each function locally, manage handoffs between them. In the agentic era that breaks entirely: velocity without flow is just a faster queue. Throughput — the number of ideas that complete the full journey from idea to monetisation — only improves when the whole system moves together. The businesses truly blooming are not the ones running multiple agents in a Ralph Wiggum loop; they are the ones adopting AI end-to-end.

Agile (capital “A”) has been dead for years. Yet, AI can be the best agile transformation you have ever experienced.

Building Smarter With AI - From Idea to Market at Warp Speed

2025-11-01T00:00:00+00:00

I was invited by Techstars and AWS to give a talk on AI to AWS Bootcamp London. Here are two key takeaways:

Adopt an AI-first mindset—design products that are AI-native and built around the unique dynamics of GenAI and agents.
Go beyond vibe coding: adopt solid engineering principles and practices so your product can adapt and scale with your business.

Slides

Building Smarter With AI-F rom Idea to Market At Warp Speed from Luca Grulla

Digital entrepreneurship and the coming “AI slop” wave

2025-04-17T00:00:00+00:00

The promise of no-code development has finally become reality through what is now called “vibe coding” — using AI tools like Lovable or Cursor, to transform ideas into functional products, just with prompts. For entrepreneurship, this shift is revolutionary: you can now go from concept to prototype in days rather than months, without any technical knowledge.

Imagine validating your business with paying customers before writing a single line of traditional code. An entire engineering team, compressed into an AI assistant, at minimal cost. The implications for innovation are staggering.

Today’s code is tomorrow liability

The Hidden Cost of AI-Accelerated Development

But as more success stories emerge, so do the challenges. Security vulnerabilities in AI-generated code are becoming increasingly documented. Hearing from people stuck in the “loop of death”, where any new prompt fix some problems while creating new ones, is common.

The “AI slop”, the term initially coined to define poor AI generated content flooding internet, is now a thing in software and product development.

As an entrepreneur, your natural optimism might suggest: “Once we find Product-Market Fit (PMF), we’ll have the resources to fix these technical issues.”

This is theoretically true; however PMF won’t give you more time — quite the opposite. Once you’ve validated your idea and raised capital from top tier investors, a new countdown begins. Investors expect rapid scaling (remember T2D3 growth? “triple, triple, double, double, double” YoY your revenue).

Nothing kills momentum faster than discovering your AI-generated foundation can’t support your growth. You’ve validated your idea and secured funding to accelerate, but you’re stuck rebuilding core systems because of accumulated debt. Your forecasted budget is suddenly challenged by this new emergency.

However, there’s a better way to embrace AI and not going blind into future success.

How to Harness AI’s Power While Building for Scale

1. The Engineering Sweet Spot: Experience + AI Augmentation

Don’t view AI as a replacement for engineers; instead, see it as an amplifier of their capabilities. Following Jevons paradox, increased efficiency driven by technology advancement doesn’t reduce demand, it increases it. You now have the possibility to do more.

Be wary of two extremes: experienced engineers who reject AI outright, and newcomers who claim to work exclusively through AI. Both miss half the equation. The sweet spot lies in combining deep technical experience with AI-augmented workflows.

2. Prioritize Data Modeling

While software is (should?) easy to change, working with data is hard. Tomorrow’s most valuable engineer will be the skilled data modeller.

Invest in defining your data architecture early—it will become the foundation everything else builds upon.

3. Double Down on Quality Assurance

In the pre AI era, a stellar engineering team might allow you to lighten testing requirements. Their competence would drammatcaily reduce the risk for trivial errors. With AI-generated code, this approach fails. Comprehensive testing becomes non-negotiable. Invest in observability tools and replay capabilities to catch issues before they reach production, and to re apply data transformation when necessary.

4. Shift left in security

AI generated code can be insecure, posing risk to your business and your customers data. Supply chain attacks, where models suggest to use non vetted libraries, are common. While I have no doubt this is an area where models will improve significantly over the years, at this stage, the moment your idea is validated, invest in security, with tools and effort.

5. Budget Realistically for AI Costs

Keep momentum required having the right budget in place. Think about your AI Expenditure then add 30-50%, per year. While cost-per-token will decrease over time, your organisation’s appetite for AI services will grow exponentially. All the services are introducing additional price tiers, to better monetise the most advanced models, the ones that you do need when building products. And we haven’t yet moved into “per AI agent” pricing, the likely next iteration of pricing.

The Balanced Approach

To thrive in this new era you can’t blindly embrace AI-only development nor stubbornly cling to traditional methods. Success will come to those who strategically leverage AI’s capabilities while building sustainable technical foundations. Because those foundations are what would make your better successful in the long term.

AI-Powered Software Creation: Trends and Strategies for 2025

2025-01-06T00:00:00+00:00

~$ sudo make my product

The emergence of Large Language Models (LLMs) is revolutionising software development at an unprecedented pace. Their generative capabilities in producing code and, on a broader scale, entire new products, are reminiscent of the Industrial Revolution. Novel methods of creating digital assets are emerging, potentially eliminating certain jobs while creating others. Combine this with disruption similar to the birth of smartphones, along with the technological innovation introduced by cloud computing, and you have a glimpse of our current era.

With all these activities and the continuous improvements in AI research, these are three expectations for 2025 and my related playbook.

Entrepreneur acceleration will require different engineering decisions

LLMs’ code/product generation capabilities will take centre stage in MVPs and prototypes. With improved time to market and less capital needed to start due to smaller engineering teams, LLMs will make entrepreneurial ideas a reality at an unprecedented pace. While it might be easy to believe we are in the “no engineer” era, I think 2025 is still the “low code” era. With the GenAI space still fluid, featuring a spread of tools and approaches, it’s critical to have a strategy to navigate decisions. My playbook for GenAI product creation for early-stage businesses would be built around two core topics:

GenAI code technology integration capabilities with human-in-the-loop. The ability to continuously hand off code/product between LLMs and humans will be key for true acceleration. Is the chosen tool a black box/walled garden, or does it natively support a human-LLM hybrid workflow?
Team expertise in “driving” decisions (instead of passively accepting LLMs decisions). While I believe smaller teams will be needed to achieve MVP stage, LLMs still hallucinate (yet less and less) and can go into an infinite bug creation loop (fix one problem and create a new one). Experienced engineers are the ones who can break the cycle and assess in real-time if the generated code is fit for the problem, making the acceleration real. We are also looking at engineers who are comfortable with the emerging style of development: conversational development with AI is here to stay.

AI platform: protocols and data contracts to the rescue

While more opportunities around AI emerge, the technology stack around LLMs is in its infancy. For example, Agentic AI is what everyone is betting on, but creating agents and agentic workflows is still complex. A set of standards is emerging, simplifying the ‘how’ and fuelling additional technical innovation at the platform level. The ModelContextProtocol introduced by Anthropic feels like a step in the right direction. In 2025, I would consider open ecosystem integration when assessing frameworks and tools. Similarly, data contracts and types will become more important than ever. For certain problem spaces, LLMs can already take over from humans. Your data contracts are critical to enable the human-AI hybrid development flow.

Future of Software Engineering: known techniques, in a new context

LLMs will shine in transforming new ideas into products, with smaller teams achieving great MVP stage, but working with mature products will still require more human expertise than LLMs. This will be particularly true for legacy systems, where internal quality can be low, making product evolution a game of whack-a-mole. LLMs can’t understand whole systems (including distributed patterns), and even less can resonate effectively on a chain of poor decisions. My headcount budgeting would be “traditional”, taking the 20% out-of-the-box productivity boost coming from AI-assisted development as a bonus. I would invest in hardening the system boundaries to facilitate incremental swaps towards the AI future.

Techniques such as TDD and pair programming will see increased adoption, with an LLM twist
- Driving expected behaviours from an LLM is most effective when combined with a TDD-style approach, involving defining the end state, iterating on prompts, refactoring, and creating tests. Pair programming with LLMs will likely become standard practice, extending beyond mere autocompletion to encompass a continuous dialogue with an AI agent throughout the feature development lifecycle. As these changes unfold, architecture’s role in designing future system growth trajectories will become increasingly central to development discussions.

I see 2025 as a transitional year, with a hybrid human-AI assisted way to create and evolve products. I’m an innovation optimist, so I expect software engineering to be completely transformed positively in the next couple of years. But which job will not be transformed by AI?

My 2023 reading list

2024-01-02T00:00:00+00:00

Another year has gone by, and I’ve selected some books about Technology, Leadership, and Innovation:

My 2023 reading list

The 9 types of Leadership: Mastering the Art of People in the 21st Century Workplace - Beatrice Chestnut
Powerful: Building a Culture OF Freedom and Responsibility - Patty McCord
Dream Teams: Working Together Without Falling Apart - Shane Snow
Fearless Change: Patterns for Introducing New Ideas - Mary Lynn Manns & Linda rising
Tidy First? - Kent Beck
Modern Software Engineering: Doing What Works to Build Better Software Faster - David Farley
The Async-First Playbook: Remote Collaboration Techniques for Agile Software Teams - Sumeet Gayathri Mogh
Edge: Value-Driven Digital Transformation - Jim Highsmith, Linda Luu & David Robinson
More Fearless Change: Strategies for Making Your Ideas Happen - Mary Lynn Manns & Linda rising

My favourite:

Powerful: Building a Culture of Freedom and Responsibility

This is one of the original books about Netflix. What struck me about that phase of Netflix is how deliberate they were in shaping and evolving the business. This book doesn’t shy away from the trade-offs they made, and it focuses significantly on the accountability piece. This book alone trumps half of the books out there on building high-performing organizations.

Is It Time to Retire Digital Transformation?

2023-10-26T00:00:00+00:00

In the webinar, I discuss the role of transformation within the enterprise and the scale-up world, arguing that the biggest transformation is at the business level, not necessarily at the technology level anymore.

Learnings from 7 years of hyper growth at Signal AI

2023-10-25T00:00:00+00:00

For the past seven years, I’ve held the position of CTO of Signal AI, a leading B2B SaaS AI platform that aids in better decision-making by analysing vast amounts of external documents. During my tenure, the business underwent significant growth. Our customer base, the scale and complexity of our ingestion pipeline, and our team size all grew by more than tenfold.

Photo by Isaac Smith on Unspash

In such an environment, change is the only constant.

And although Agile and modern product development principles embrace change, the realities we face are not the neat iterations where technology teams ideally flourish. It’s much messier than that.

As the CTO, I realised that while certainty might not be achievable in scaling businesses, most of my work centred around providing as much clarity as possible and foster an environment resilient to uncertainty.

Here are my three key learnings:

A technology strategy is your compass to balance reactive and proactive work

I firmly believe that in scenarios with an in-house technology team, merely responding to changes is insufficient. You need to anticipate changes and proactively and incrementally put things in place so that when the time comes, you can perform rather than form. This proactive approach is even more crucial in B2B SaaS, where the feedback loop between initiatives and outcomes is much longer and less direct than in D2C. A technology strategy serves as the compass to traverse changes and underscore the importance of long term initiatives. It’s also an essential tool for the broader business, acting as a barometer to gauge how emerging opportunities weigh against long-term objectives. The strategy is not your roadmap or a detailed technology selection plan, but a high-level framework that should enable people to make relevant decisions.

Design for resilience

In such a fast-changing environment, context can shift with little to no notice. Factors like accelerated customer adoption or changes in usage can exert pressure on unforeseen system components. Designing a failure-tolerant system is key to absorbing these shifts: controlled service degradation, will allow you to manage the unexpected while going back to the drawing board to plan the next step. A modern stack and healthy foundational engineering principles are all you need. Highly cohesive and lowly coupled systems, good integration principles (pro tip: Microservices is one of - and not the only - solution), run on an elastic platform go a long way. Investing in observability is key, so you can understand what’s happening when things go off rails. It’s critical to match the technical solutions with the right mindset: it’s a shift from having a list of SLAs to hit to a way of thinking and designing software. Counterintuitively, this mindset will not slow down your time to market; by having people more aware of the low-hanging fruit of designing failure-tolerant systems, you will enable faster prototyping principles, and better time to market.

You might never achieve the ideal org design

While in hyper-growth, your rate of change is higher than the adoption rate from teams and people. Teams are more often in the form/storm stage than the perform stage. Use team topologies as a north star, but acknowledge that it’s an ever-evolving scenario. To balance out decisions, I kept assessing individual team maturity and the pace of change of each domain. The domains with more business and technical complexity, tend to be more resilient to change. That’s because, due to their complexity, their pace of change is slower. When possible, use those teams as a springboard to the rest of the organization. Be honest with people and teams about the journey ahead, and how things might change or how their contribution to onboard new team members (at the cost of some team stability) is part of their mission.

Communication and great people are the glue to all these. Emphasise talent density over quantity, treat people like adults, and encourage honest and open communication.

More than everything else, it’s a crazy rollercoaster; celebrate achievements, and don’t get too low on the inevitable mistakes along the way.

My 2022 reading list

2023-01-03T00:00:00+00:00

Reading is, by far, my favourite way to learn and reflect on ideas. The themes I invested more in over 2022 include leadership, technology, organizational design and company culture.

Other books on the same themes are in my 2020 reading list and 2021 reading list.

My 2022 reading list

The Messy middle - Scott Belsky
The 4 Disciplines of execution - Chris McChesney, Sean Covey, Jim Huling
The power of a positive no - William Ury
The art of Leadership: Small things, done well - Michael Loop
The ultimate data and AI guide - Alexander Thamm, Michael Gramlich, Alexander Borek, Ronald van Loon
The fearless organization - Amy C. Edmondson
Decisive - Chip Heath, Dan Heath

These two books stood out on the list:

The Power of a Positive No

Behind every no, there’s a yes, something we are protecting, or other things we are willing to say yes to. The ability to explain the reasons behind a no, together with an alternative is critical to move forward and achieve a meaningful agreement. And This book is a gem.

The Art of Leadership

One of the best books about Technology Leadership I’ve ever read. Many books focus on the practices and activities; this gives insights into the why, the mindset, and the tradeoff you need to take. If you are in a technology leadership role, get this book.

My 2021 reading list

2022-01-03T00:00:00+00:00

As I did for 2020 in this post, the end of the year is an opportunity to share the books I’ve read over the last 12 months.

My 2021 reading list

These two are the ones that gave me more food for thought:

No rules rules

A lot has been shared about the Netflix culture, often with an accent on the high degree of autonomy and decentralized decision making. But what is required to get the benefit of such a model? The book from Netflix’s CEO does an excellent analysis of what Netflix culture truly means and the trade-offs needed to make it work. A great read.

Team Topologies

This book has quickly become a pillar for Technology leaders. After reading it, I changed how I translate strategy to org design, the language I use and how I help teams find their missions and goals.

Preparing for a Technical Due Diligence

2021-12-20T00:00:00+00:00

CircleCI invited me to a panel on Technical Due Diligence, part of their “Startup Spaces” series. It was a great conversation, with a healthy mix of Auditors and CTOs on the roster.

These 4 tips are the summary of my sharings.

Understand the big picture

Understand the investment thesis, and capture how the technology will support and accelerate success. Is it about geographical expansion? Or is it about launching a new product? Or maybe, is about improve end to end business efficiency? Your future technology state should point in the same direction as the business.

It’s a story about the technology opportunity

You are not just sharing the characteristics of your tech stack or your process. You tell a story about the optionality created by technology; you show how technology accelerates the business to the next growth stage. Use the relevant metrics to support your message, and focus on your strength. IPs and patents are a great way to demonstrate defensibility and intrinsic value.

Think about it at 360 degrees

Although it’s easy to focus on architecture and process, cross-functional requirements are a critical part of your DD. Security, data privacy, disaster recovery are essential topics, and as a CTO, it’s your job you have an answer. These themes get more and more scrutiny the more you move through your investment rounds, and so your organisation sophistication should increase over time.

Give yourself time

Start thinking way ahead about areas that require some attention in preparation for a funding event. For example, cross-functional requirements might need time to be polished.

Prepare some collaterals ahead of time; architecture diagrams, metrics on uptime, Disaster Recovery summary will be handy. Having them ready will give you confidence and avoid the pressure of preparing them in a rush.