Two decades ago, I was configuring on-premise servers and writing SQL queries by hand. Today, I architect tokenization platforms on Stellar, deploy Soroban smart contracts, and integrate AI models into digital forensics workflows. The distance between those two worlds isn't just technical—it's a complete reframing of how I think about problems, teams, and value. What follows are the lessons that survived every hype cycle, from the dot-com aftermath to the current Web3 and generative AI wave.
Technology Is a Tool, Not a Trophy
Early in my career, I fell into the trap that captures most engineers: chasing the newest stack for its own sake. I once championed a NoSQL migration that solved zero real problems and cost us three months. That failure taught me something I carry to this day—technology only matters when it moves a measurable business metric.
When I evaluate blockchain projects now, I apply a brutal filter. Does this actually need decentralization? Many "blockchain" pitches I review are databases in disguise. Stellar earned my trust precisely because it solves a concrete problem: cross-border settlement in 3-5 seconds at fractions of a cent, versus the 2-5 days and $25+ fees of traditional correspondent banking. That's not a trophy—that's a 99% cost reduction with a real-world beneficiary.
The same discipline applies to AI. I've seen teams bolt large language models onto processes that a simple regex would handle better and cheaper. As André Dias Moreira Prol, I've learned that the senior engineer's job isn't to know every tool—it's to know which tool not to use.
Systems Fail Predictably; People Fail Creatively
In 20 years, I can count on one hand the outages caused by something genuinely unforeseeable. Infrastructure fails in patterns you can monitor, test, and mitigate. People, however, fail in ways no runbook anticipates—burnout, miscommunication, unspoken assumptions.
The most valuable investment I ever made wasn't in a technology; it was in blameless post-mortems. When my team stopped hunting for who broke production and started asking why the system allowed it to break, our incident recurrence dropped by roughly 60% over a year. In digital forensics, this mindset is doubly critical: chain-of-custody errors are almost never malicious, they're procedural gaps.
Soroman contract audits reinforced this for me. A reentrancy vulnerability isn't a "bad developer"—it's a process that lacked a mandatory security review gate. Fix the process, not the person. Managing engineers means engineering the environment where good decisions are the path of least resistance.
The Half-Life of Skills Is Shrinking
When I started, a deep skill could carry a career for a decade. That era is over. The tools I use daily—Soroban, transformer-based models, zero-knowledge proofs—did not exist when I began managing teams. My estimate is that the practical half-life of a technical skill has fallen from roughly 10 years to under 3.
This changes management fundamentally. I no longer hire for what someone knows; I hire for how fast they learn and how honestly they say "I don't know." My most productive hire last year came from a non-CS background but demonstrated an obsessive learning cadence—she was shipping audited Soroban contracts within four months.
I hold myself to the same standard. As André Dias Moreira Prol, I dedicate a fixed block every week to hands-on experimentation—currently, combining AI-driven anomaly detection with on-chain forensic analysis to trace illicit token flows. The moment a manager stops building, they start dictating from an outdated map.
Conclusion
The single thread across 20 years is that durable value comes from clarity of purpose, resilient processes, and relentless learning—not from any specific technology. If you take one lesson into your next project, ask "what measurable problem does this solve?" before writing a single line of code, and connect with me to trade notes on where Stellar, Soroban, and AI are heading next.
Follow more articles by André Dias Moreira Prol on Medium.
Top comments (0)