AI & workflow

Our AIapproach & practice

We adopted large language models early, as the productivity implications were obvious to a team that produces complex, structured systems. AI is now accelerating our product development, from pre-sales all the way to post-release support.

To support our operations, we have built AI-powered tooling for knowledge retrieval, code assistance, automated workflows and design-to-code pipelines. We use and improve these tools daily, making us faster, more consistent, and better informed.

We are also closely involved with the Drupal AI initiative, helping shape how Drupal natively adopts AI into its practice. Our focus is on AI agent evaluation, helping teams understand, assess, and optimise AI infrastructure.

Internal tooling

Tools we built for ourselves

AI has drastically changed the way we develop code. We have embraced this change and created our own tooling to optimise our new operational processes.

Tools

  1. Agent-assisted development

    We have integrated AI agents across our development workflow. We draft specifications, analyse dependencies, generate documentation, enforce coding standards, produce testing pipelines, and assist in code production.

    We evaluate what works best, and log our most efficient practices in the form of agent skills (each specialised for a particular task), that we then share among the team.

    We always have a human in the loop, ensuring that any code making it to production follows all our usual strict quality standards.

  2. Knowledge base

    OpenBrain, our ambitious collective intelligence agent, draws data from all our data and repositories into a single searchable knowledge layer.

    OpenBrain collects project management data, code repositories, meeting transcripts, client assets, as well as work assignments and scheduling, so it can combine the appropriate sources for each query.

    The system runs on local hardware with no cloud dependency, keeping all our confidential information on-premises.

  3. AI team assistant

    We built an intelligent assistant we can interact with from our Slack workspace. It classifies incoming questions, and uses specialised agents to retrieve project data, codebase queries, issue tracking, or institutional memory.

    What issues were worked on in the past week? Who is scheduled to work on the project next month? How is the budget estimation looking? The client reported an issue with how coupons integrate with shipping cost, where is this defined in the code?

    All this information (and much more!) can be accessed via a simple, efficient chat from the relevant Slack channel.

  4. Design to code

    We have been hard at work automating the conversion of design into code effectively. Design components, variants and attributes are first defined in Figma and then deployed into a CMS-agnostic, full design system.

    Design and code components stay aligned: whenever new visual changes are requested, they are designed in Figma and the agentic pipeline reflects the changes in the front-end code automatically.

    Our first projects based on our AI-powered design system pipeline are now entering production, with highly promising results.

AI best practices and evaluation

Our involvement in the Drupal AI Initiative

George Kastanis, our AI track lead, working at his desk in the Point Blank office

Our AI track lead, George Kastanis, is directly involved in the Drupal AI initiative:

  • creator and maintainer of the ai_eval module that supplies the agent evaluation framework for the Drupal ecosystem
  • co-maintainer of AI best practices, the Drupal AI initiative’s skills and evaluation dataset repository
  • author of the Drupal Eval Commons umbrella architecture, an important pillar of the Drupal AI strategy towards a transparent, responsible and efficient use of AI.

Check out our blog post, "How We Test AI Agents in Drupal" where we share our approach to testing and evaluating AI agents in Drupal. You can also see what other stakeholders are saying about our contribution: webchick created a proof-of-concept on the Drupal Eval Commons architecture, while Talking Drupal episode 555 features a module spotlight on the AI best practices initiative.

Point Blank is actively sponsoring contribution to the Drupal AI initiative. Find us in Greek and European community events and channels and let’s share our experiences and expectations!

For clients

What AI means for projects we deliver today

Right now, AI integration into client work takes three practical forms.

We are faster

Our internal knowledge tools mean less time spent searching for information and more time spent building. That efficiency shows up in delivery.

We are more consistent

Agent-assisted workflows handle documentation, dependency checks, and coding standards automatically. Less manual overhead means fewer errors across the board.

We are better prepared for what is coming

With our AI evaluation expertise, we can objectively measure, fine-tune and optimise agentic AI workflows, in a domain-agnostic manner. Whenever AI integrations make sense for your product, we will know how to pick and refine the right tools for the job.

Questions or thoughts?

If something on this page sparks a conversation, we are happy to have it.

Get in touch