Skip to content
    talking agents.

    Specialist engineering partner · Finland

    Production engineering for agentic AI systems.

    From impressive demos to reliable workflows.

    We help AI teams build the architecture, MCP integrations, permissions, evaluation loops, and human approval paths that production demands.

    Book a technical scoping callExplore the work

    Hands-on engineering by Talking Agents Oy.
    Based in Espoo, Finland. Working with teams everywhere.

    01 / System boundariesReference workflow
    A clear path from intent to action.
    Boundary 03

    Pause consequential actions for a named reviewer. A rejection returns the task for revision; no approval means no execution.

    Explicit ownership at every step

    Built for your next step

    • AI startups
    • AI consultancies & agencies
    • SaaS teams building agents
    • Teams adopting MCP

    01 / How we can help

    A concrete scope.
    A working result.

    Start with the part that is holding you back. Each engagement has an agreed scope, clear deliverables, and a handover your team can use.

    01Agent System Architecture ReviewBefore you commit to the next build.

    Review your agent design, context boundaries, tool access, and failure modes against the workflow you need to ship.

    What you leave with

    An architecture map, a prioritised risk register, and an implementation plan.

    Discuss this engagement
    02MCP / Tool Integration SprintConnect agents to the systems your team actually uses.

    Build scoped tool integrations with explicit identity, permissions, input validation, and useful error handling.

    What you leave with

    Working integrations, contract tests, and integration documentation.

    Discuss this engagement
    03AgentPlanner Deployment & Workflow SetupGive multi-step work a shared structure.

    Deploy AgentPlanner in your environment. Connect MCP clients and configure plans, dependencies, and human review points.

    What you leave with

    A configured deployment, a first working workflow, and a team handover.

    Discuss this engagement
    04Production Hardening SprintWhen a successful demo still needs supervision.

    Work through evaluations, traces, timeouts, retries, and recovery. Make failures visible and define how the system gets back on track.

    What you leave with

    An evaluation baseline, operational visibility, and a recovery runbook.

    Discuss this engagement
    05AI Workflow Governance SetupMake authority and accountability explicit.

    Define who owns each workflow, what agents may access, which actions need approval, and what evidence is recorded.

    What you leave with

    A permissions matrix, approval paths, and a documented ownership model.

    Discuss this engagement

    The engineering perspective

    The hard part
    is not prompting.

    It is context, permissions, reliability, ownership, and knowing when a human must stay in the loop.

    Bounded authorityVisible failuresHuman accountability

    02 / Selected work

    Engineering experience.
    Something you can inspect.

    Client work in industrial ML and gaming. Open source agent infrastructure you can explore for yourself.

    Our own product / Open source

    AgentPlanner

    We build AgentPlanner: a shared planning and coordination layer for agents and the people responsible for their work.

    • Structured plans and task dependencies
    • Persistent context across agent sessions
    • MCP connections and human review points

    Use it independently, or work with us to deploy it and build your workflows.

    Inspect the source Explore AgentPlanner
    AgentPlanner release plan showing task completion, a plan tree, and dependency indicators.
    Product view / Release planPlans, progress, dependencies.
    Industrial operations / Machine learning

    Predictive maintenance

    Machine learning work for industrial predictive maintenance, bringing software engineering and modelling into an operational setting.

    Industrial ML
    Gaming / Machine learning

    Recommendation systems

    Recommendation system work in gaming, applying machine learning to how products surface relevant choices to their users.

    Recommendation engineering

    03 / Your engineering partner

    Specialist expertise.
    Shared responsibility.

    Talking Agents is a specialist engineering company building production agent systems and ML workflows.

    Based in Espoo, Finland, we work alongside your team from architecture through implementation and handover. Our work connects software engineering, applied ML, and the practical experience of building AgentPlanner.

    Founded by Michael Malka, whose background includes over 20 years in software and technology leadership.

    About Talking Agents

    Start with a conversation

    What stands between
    your demo and production?

    Bring your workflow, architecture, or the problem you keep running into. We’ll use 30 minutes to identify the engineering work and see whether we’re a good fit.

    Book a technical scoping callPrefer email? contact@talkingagents.com
    A useful first conversation
    1. 01What needs to work?
    2. 02Where does it break today?
    3. 03What is the smallest useful next step?