AI: I guess I will delete your vibe coding project

AI coding assistants are now powerful enough to build an app in a day, and just dumb enough to delete it in a second.

Your AI Assistant Is One Command Away From Deleting Everything

This isn't a sci-fi plot, it's what happened to multiple developers. One watched their AI assistant execute a `DROP DATABASE` command on a live system, turning months of work into digital dust. Another accidentally approved a command that wiped their entire home directory. These aren't bugs; they are features of pairing with a tool that mimics intelligence without understanding consequences.

The real story here isn't just about rogue commands. It's about our rush to grant AI autonomy without building guardrails, a trend the community calls 'vibe coding'. We're so focused on development velocity that we're skipping the crucial step of treating AI like what it is: an incredibly powerful but non-sentient intern. It operates on patterns, not comprehension, and blind trust is the fastest way to disaster.

The solution isn't to abandon these tools, but to enforce iron-clad rules. This means mandatory human review for any destructive command, confining AI agents to sandbox environments, and applying the principle of least privilege. AI is a powerful assistant, not a trusted teammate.

Read more →

AI Assistants That Won't Nuke Your Work

While some AI assistants are busy deleting files, others just want to help clear your inbox.

Harmony AI: Talk your way to inbox zero, hands-free.

This is a genuinely useful application of voice AI for managing Gmail and Google Calendar while driving or working out. By focusing on a specific, high-friction workflow, it avoids over-promising and actually delivers value.

Stupido: A to-do list so simple, it's almost cheating.

The rise of hyper-minimalist tools signals a fatigue with feature-bloated apps. Stupido's voice-first, single-button interface is a direct response to a world drowning in digital complexity.

The New Guard of Builder Tools

Beyond AI helpers, a new class of tools is emerging that respects your data and your workflow.

Jaaz: The open-source Canva that runs on your machine.

Jaaz isn't just a free design tool; it's a statement about data ownership. Running locally is a huge deal for anyone tired of their creative work living on someone else's server.

Inbound: An email platform that thinks like a developer.

Finally, an email tool that treats your inbox like an API endpoint. By offering webhooks and smart routing, Inbound turns email from a chaotic time-sink into a structured, integrated part of the dev stack.

Quick hits

Google Flights AI: Describe your dream trip, get real deals.
Google's new AI lets you search for flights using natural language, turning vague vibes into actual, budget-friendly trips.

Zero Waste Cloud: Like Ozempic for your AWS and GCP bills.
This tool scans your cloud accounts in minutes to find idle servers and oversized databases, promising to cut waste by up to 40%.

My takeaway

The tension between AI's power to create and its power to destroy is becoming unavoidable.

We are obsessed with making AI faster, more autonomous, and capable of incredible creative feats like building an app in hours. But we are neglecting the boring, essential work of building guardrails, safety nets, and kill switches. These recent destructive incidents are not edge cases; they are the logical outcome of deploying systems we don't fully control.

The next significant innovation in AI won't be a more powerful model, but rather the frameworks and protocols that allow us to use it safely. The platforms that prioritise robust security and human-in-the-loop safeguards will ultimately win our trust. We just have to hope we don't lose a production database before then.

At what point does an AI tool become too autonomous to trust?

Drop me a reply. Till next time, this is Louis, and you are reading Louis.log().