Levantar AI

New research worth paying attention to if you're building with LLM-generated code 👀 A team of researchers has released LLMloop, a framework designed to automatically refine code and test cases produced by large language models through five iterative feedback loops.

Thrilled to welcome Highfield Qualifications to Upcast 🎉 Applied AI that delivers real value — not AI for its own sake. Excited to build this partnership. highfieldqualifications.com | upcast.social

McKinsey found orgs with centralised AI governance are 2.5x more likely to see real productivity gains. The bottleneck isn't AI's power — it's invisible workflows and ungoverned adoption. 🤔 ability.ai/blog/ai-workfl…

Here's something that keeps coming up in our architecture reviews, and it's worth talking about openly 🔍 Getting an AI feature to work in a demo takes days.

"The faster the car, the better the brakes you need." 🏎️ AI adoption is outpacing governance structures built for slower development cycles.

New research from the Technical University of Denmark and The LEGO Group examines how smaller, open-source language models can be customised for domain-specific code generation — and the results are worth paying attention to 🔍 The study, led by Luís Freire, Fernanda...

The World Economic Forum has published a new report — "Organisational Transformation in the Age of AI" — drawing on insights from over 450 executives in its AI Transformation of Industries community 📊 The core message: AI has moved past experimentation.

The agentic AI market is projected to grow from roughly $5.2 billion in 2024 to $52.6 billion by 2030, according to MarketsandMarkets. That's close to a tenfold expansion in six years 📈 A recent analysis from The Motley Fool breaks down how this growth is playing out in...

New research from Linköping University highlights a gap we think deserves more attention: the non-functional quality of LLM-generated code 🔍 The study — a mixed-methods investigation covering 109 papers, industry workshops, and empirical analysis using three LLMs on...

SlashData just dropped a packed roundup of 14 developer trends heading into Q1 2026, and there's a lot worth digging into 👇 Here's what stands out: 📊 There are now 48.4 million developers worldwide.

Here's a question that doesn't come up often enough when teams start building AI agents: who handles failure? 🤔 Not the model. Not the orchestrator.

Fascinating new research that puts LLM code generation capabilities into sharp perspective 🔬 Researchers Aman Sharma and Paras Chopra have introduced EsoLang-Bench, a benchmark that tests large language models against esoteric programming languages — think Brainfuck,...

2026 is shaping up to be the year AI agents move from pilot programmes to core enterprise operations 🚀 A recent piece from AppZen pulls together predictions from Gartner, Forrester, PwC, and Deloitte — and the consensus is striking: 📌 Gartner forecasts AI agents...

⚡ Iran's rhetoric toward the US is sharpening. When geopolitical tensions rise, nation-state cyber activity follows — especially across supply chains. Startups: if you touch defence, energy, or finance, even tangentially, this applies to...

Here's a pattern we keep seeing in AI-augmented development workflows that's worth calling out 🔍 Teams adopt Copilot, Cursor, or similar code-generation tools. Productivity spikes in the first few weeks. Then it plateaus — sometimes even dips.

£49/month or 3 hours of engineering time maintaining a DIY honeypot? 🛠️ Tripwires gives you 10 protocol sensors — deployed from a dashboard, not a terminal. Start free with unlimited decoy docs 👇 tripwires.levantar.ai

The Pragmatic Engineer just published findings from a survey of 900+ software engineers on AI tooling in 2026, and the data is striking 📊 Here's what stands out: Claude Code went from zero to the most-used AI coding tool in just eight months since its May 2025 release,...

Coding LLMs in 2026 aren't competing to be "the best" — they've specialised. GPT-5.2 Pro for logic, Claude 4.5 Opus for architecture, Gemini 3 Pro for 2M-token context, Llama 4 for private deployment. But models still hallucinate fake libraries with total confidence 😬

SSH honeypot on your network = instant alert the moment someone tries credentials that don't exist. No false positives. No tuning. £49/month gets you 10 of these sensors. That's less than a team lunch. 🍕 tripwires.levantar.ai

Lateral movement leaves traces — if you've got something watching. 👁️ Tripwire Documents sit quietly on endpoints and file shares. £0, no registration, unlimited. The moment someone touches one, you get source IP, location, and timing. tripwires.levantar.ai