This edition focuses on how enterprise teams build and scale AI in real systems - across engineering partnerships, agent workflows, model evaluation, and production infrastructure.
Inside, we share how our work with Microsoft and Reddit has expanded to include data, ML, and AI; what it takes to scale AI agents in fintech with compliance and measurable ROI in mind; how AI models perform in GitHub Copilot CLI; and what engineering judgment looks like under tight constraints.
We also bring key signals from TECHARENA Stockholm and WAICF Cannes on how enterprise teams are approaching AI adoption today.
Microsoft & Reddit on Working With Akvelon
We’re proud to share how leaders at Reddit and Microsoft reflect on what it’s like to work with Akvelon – watch the video.
At Microsoft Azure, Michael Lu, Principal Software Engineering Manager, shares how the collaboration has grown from UI work into broader support across data pipelines, ML, and AI. At Reddit, David, Engineering Manager, explains how Akvelon’s engineers contribute to core platform systems, build integration tools, improve conversion tracking, and strengthen system reliability. Microsoft’s Principal PM Manager, Henry Dixon, highlights how engineers onboard quickly, contribute across teams, and help scale delivery.
We’re grateful for the opportunity to support this work, and for the trust these teams continue to place in Akvelon. If you’re building complex systems or scaling AI, we’re ready to support your team –> reach out.
Why Subagents Fail Without Context Isolation
Subagents are often treated as different roles working together. But roles alone don’t solve the problem. Our Director of AI/ML Engineering, Ilya Polishchuk, explains why context isolation is the real value in his article.
When one agent holds too much context — requirements, code, logs, prior reasoning — noise increases and quality drops. Splitting work into clean, bounded contexts improves focus and reliability.
Choosing the Right Model for AI Code Review
Model choice directly affects how useful AI code review is.
Our Director of AI/ML Engineering, Ilya Polishchuk, benchmarked several models using the GitHub Copilot CLI on Medium.
The results show how differently models perform at finding real issues without adding irrelevant noise:
- Tested on 50 real pull requests using a public benchmark
- Evaluated using precision, recall, and F1
The article also covers key limitations, including dataset bias and variability across runs, and explains how to evaluate models in your own workflows.
If you’re building AI-assisted development workflows, this material gives you a clear way to test and choose the right model.
AI Agents in Finance: What It Takes to Scale Safely
AI agents don’t deliver ROI in finance without the right foundation. Our whitepaper outlines how to deploy them safely in regulated environments.
What's inside:
- Where AI agents reduce costs first, across onboarding, fraud, reporting, and operations
- How compliance and audit logic are baked in from day one
- What production-ready deployment looks like
- How teams measure ROI beyond the pilot stage
For teams scaling AI in finance, this whitepaper breaks down how to build systems with the control, visibility, and oversight required in production.
The 1-Hour Code Challenge: Testing AI Under Constraints
How do strong engineers work through a tightly scoped task when the inputs are messy, and the clock is running?
In Akvelon’s latest article, that question is explored through a practical internal exercise: one well-scoped real-world case, 60 minutes, predefined conditions, and a hard deadline.
The focus is not on building a project in an hour, but on what drives results on a bounded coding task: problem framing, assumptions, iteration, verification, trade-offs, and judgment under pressure.
Join Us at These Upcoming Events!
Kirill Nesterenko, Akvelon’s Director of New Business Development, will be attending HANNOVER MESSE 2026 – the world’s leading trade fair for industrial technology on April 20–24 · Hannover Exhibition Grounds, Germany
Ashley Pikle, Akvelon’s Director of AI Business Development, will be attending:
- Google Cloud Next ’26: Google Cloud’s flagship event focused on cloud, data, and AI April 22–24 · Mandalay Bay Convention Center, Las Vegas
- Momentum AI: Event focused on practical AI adoption and enterprise use cases April 27–28 · Austin, TX
- TechEx North America: Conference covering enterprise technology, including AI, cloud, and data May 18–19 · San Jose McEnery Convention Center, CA
- AI Dev Summit: Conference for AI engineering, ML systems, and production AI May 27–28 · San Francisco, CA
If you’ll be attending any of these events or would like to connect, reach out to Kirill and Ashley via LinkedIn.
Signals from Recent Events: TECHARENA & WAICF
At WAICF Cannes 2026, Kirill Nesterenko, Director of New Business Development, focused on how enterprise AI is being deployed at scale. What stood out was how far things have moved into production. Teams are already running agent-based workflows, with governance, compliance, and auditability treated as core parts of system design. AI is increasingly being deployed in business-critical, regulated environments, where systems must operate reliably under real constraints.
At TECHARENA Stockholm 2026, Kate Nyzhehorodtseva, Director of New Business Development, saw the same shift play out in a different context. AI is no longer treated as an experiment. It’s becoming part of long-term infrastructure. Teams focus on measurable outcomes, integration into existing systems, and embedding agents directly into workflows. AI isn’t something added later anymore. It’s now built into system design from the start.
Keep up with the latest tech news, carefully curated and analyzed by Akvelon's experts:
- Microsoft: Building the Next Layer of AI Infrastructure: Microsoft is bringing NVIDIA’s Vera Rubin NVL72 systems into Azure for validation, another step toward next-generation AI infrastructure. For enterprise teams, this reinforces a shift toward infrastructure-first AI. As systems scale, success depends less on model choice and more on how workloads are integrated, deployed, and operated across platforms.
- Google Cloud: From Agents to Real Workflows: Google Cloud released a developer guide with frameworks and code samples for building production-ready AI agents. The key question is no longer what agents can do, but how they fit into existing workflows and systems.
- NVIDIA: Cost and Scale Become the Priority: NVIDIA highlights new large-scale deployments, including Meta’s infrastructure build and significant cost reductions achieved with open models on Blackwell. At scale, AI turns into a cost equation. Cost per inference, performance trade-offs, and infrastructure choices define what’s viable in production.
- Microsoft: Optimizing for Inference Efficiency: Microsoft introduced its Maia 200 accelerator, designed for better performance per dollar and large-scale AI workloads. As AI systems grow, inference efficiency becomes a key design factor, shaping system-level cost, latency, and reliability.
Share your feedback and ideas about what you’d like to see on Akvelon’s LinkedIn by emailing info@akvelon.com.