I Fixed the Dumbest Thing About AI Agents
How I built a shared, persistent memory layer across four AI-agent runtimes to reduce repeated briefings and keep decisions consistent between sessions.
Read the article00A PERSONAL LABORATORY / ASCENDANCY IQ
Work is becoming
agentic.
The question is what we do with it.
I’m building an autonomous working system that gives people the capacity to do more. A workforce organised around real work, shared context and business outcomes.
Enter the story BUILT BY KEITH DE ALWIS · EVOLVING IN THE LAB01THE PROBLEM
More intelligence.
More noise.
Another model. Another tool. Another conversation starting from zero.
There is extraordinary capability everywhere. But joining it up still falls to us. We carry the context, chase the handoffs and turn a hundred promising outputs into something useful.
DISCONNECTED TOOLS. FRAGMENTED CONTEXT. HUMAN GLUE.02FROM INTENT TO ACTION
One conversation.
A workforce
in motion.
Can I agree the priorities, then trust the work to move forward?
That is the question behind AIQ. Talk it through with an orchestrator. Agree the outcome. Turn it into owned, trackable tasks. Let specialist employees work in parallel, bringing decisions back when they need me.
DISCUSS → AGREE → DELEGATE → DELIVER03BUILT AROUND THE WORK
Different skills.
One shared
purpose.
A digital employee needs more than a clever prompt.
I personally shape each employee around outcomes, responsibilities, skills and tools. The engineering lab began with orchestration, technology and design. The ambition extends across research, growth and the everyday work of a business.
PERSONALLY SHAPED. CONTINUALLY TESTED.04INSIDE THE SYSTEM
The architecture
is designed to
outlast the models.
Intelligence is one layer. Making it useful takes a system.
Human direction, a shared control plane, specialist employees, persistent knowledge and connected tools. Underneath, an execution layer designed for frontier and open-weight models, in the cloud or on physical hardware.
Explore the architecture A HIGH-LEVEL VIEW · ARCHITECTURE EVOLVING IN THE LAB05CONTINUITY
Work should not
begin from zero
every morning.
The next session should know what the last one learned.
AIQ Brain gives the workforce a shared knowledge repository. Decisions, runbooks, project state and lessons from real work stay available across sessions and runtimes. Context becomes something the system can carry forward.
Read how I built the brain DECISIONS. KNOWLEDGE. RUNBOOKS. CONTINUITY.06AUTONOMY WITH ACCOUNTABILITY
Autonomy needs
a way to
say no.
More freedom to act needs clearer boundaries.
Signed operations, scoped permissions and audit trails underpin the work. Human approval belongs at the decisions that matter. A capable system should know when to proceed, when to ask and when to stop.
CLEAR AUTHORITY. VISIBLE WORK. HUMAN JUDGEMENT.07A LABOUR OF LOVE
Built for my work.
Tested against
the real world.
I am the first person who has to live with the results.
AIQ is a working laboratory, built and tested through my own work. It draws on open source and lessons from dozens of repositories and projects. I’m finding what holds together, what breaks and what deserves to become a controlled beta.
LAB PROJECT · WORKING TOWARDS A CONTROLLED BETA08THE DIRECTION OF TRAVEL
One person.
The right workforce.
Far more possible.
Built for me. With the ambition to make it useful to others.
My vision is a world where one person, supported by the right AI workforce, can build, market and scale a business that once needed a much larger team. The human sets the direction. The system helps carry the work.
Follow the build A VISION BEING BUILT, ONE REAL OUTCOME AT A TIME.FROM INSIDE THE LABThe work.
The work.
And the thinking.
The experiments, the awkward bits and what I’m learning along the way.
The architecture, in plain sight
The principal
Goals, priorities and judgement remain human. Agree the outcome, set boundaries and approve consequential decisions.
Coordinate the work
Turn intent into tasks, route work to the right employee and bring progress, exceptions and approval back into one place.
Specialists with ownership
Employees have defined responsibilities, skills, tools and expected outcomes. The engineering lab pairs orchestration with technology and design.
Context that persists
A shared, versioned knowledge repository carries decisions, runbooks and project state across sessions and agent runtimes.
Connect to real work
Tool connections and signed bridges link the workforce to business systems, with scoped access and an auditable record of operations.
Choose the right runtime
A model-agnostic, node-based approach supports frontier and open-weight models and work across cloud and physical hardware. Deployment capabilities continue to evolve in the lab.
This is a high-level account of a working lab, including its design direction. It is not a claim that every capability is ready for general release.
THE LABORATORY IS OPENCome along
for the build.
I share what I’m making and what I’m learning in The Sunday Blueprint.
Follow the build Talk about AIQ