Cleverbit Software

When an AI agent writes the code, who is accountable for the decisions inside it? For years, the answer was simple. You ran git blame, found who committed the change and asked them why. With agents now making decisions of their own, the commit no longer tells you who decided what. Join our CTO Eman Zerafa and AI Solutions Engineer Ethan Zammit on Friday 16 October at 3PM CET for "When the agent decides, who answers? Accountability in AI-assisted delivery." They'll cover: *Why code review can't carry accountability for AI-written code *The controls teams lost without noticing, and how to rebuild them *A practical model for deciding what to automate and where human judgement belongs Built for CTOs, VPs of Engineering and engineering leaders scaling AI adoption. Register free via Zoom.

Most agentic AI failures are architecture failures, not model failures. Give an agent write access to a pipeline and enough autonomy, and it will eventually do something you didn't plan for, so the real question isn't whether that happens but whether you can see it when it does. Our new post breaks down what governance looks like at the architecture level: an autonomy envelope that separates agentic decisions from human-owned ones, version control for every agent-facing artefact, pipeline-level enforcement instead of after-the-fact checks, and a pilot-first rollout that scales on evidence, not enthusiasm. Link in comment 👇 #AgenticAI #SoftwareArchitecture #AIGovernance #CTO #SDLC

Every day, employees paste proprietary source code, confidential contracts and sensitive financial data into public AI tools to save time. What most of them do not realise is that once that data crosses your digital perimeter, it is no longer yours. It can be stored, reviewed and in some cases surfaced to competitors. There is a name for this: shadow AI. It is not malicious. It happens because your official tools are too slow and your team is trying to hit their targets. Episode 5 of Optimise covers the three pillars of enterprise AI security every CEO needs to have in place: a zero trust data perimeter, contextual access control and an automated verification layer. Watch episode 5 here: https://lnkd.in/d_nQNvfB

Ever asked an AI agent for something simple and got something else back? We call it vibe code drift, and it's becoming one of the biggest risks in AI-assisted delivery. Download our free Field Guide to see how to catch it early.

Most companies think they have an AI strategy because their team uses a chatbot every day. They don't. They have a faster search engine ;-P A chatbot is reactive. It sits there waiting for a human to type something before it does anything. Autonomous agents work differently. You give them a goal, a set of tools and guardrails, and they execute. Episode 4 of Optimise covers what autonomous agents are, the four pillars you need to design one properly and why this cannot be built on a $20 per month subscription. Watch it here: https://lnkd.in/dG6vKZ-a

One phrase changed everything. Same request. Same model. One prompt included "keep it short and lightweight." The other didn't. With that phrase: Claude gave us a single line of code that shipped with an XSS vulnerability. Without it: Claude caught the vulnerability and wrote secure code. This is what we call the salience trap. AI focuses on whatever's most salient in your prompt. Everything else gets ignored—even security. This is one example from our webinar on agentic delivery. We walk through the patterns that keep recurring: how one instruction overrides another, how agents fill gaps you didn't know were there, why verification is now the bottleneck, and what it takes to stay in control. Watch the short. Then watch the full webinar. It's 45 minutes that change how you think about prompting. Link in comment