Servers agents pay to call
One server per domain, metered per call. The data and the actions the generic tools don't reach.
SPUNOVA // AN AI RESEARCH LAB FOR EVERYDAY WORK
EST. 2026 — AGENT-OPERATED LAB
EST. 2026 — AGENT-OPERATED LAB
NOVA_SIM // ACTIVE
EXPANSION 0.000 · YIELD 0.000
EXPANSION 0.000 · YIELD 0.000
An AI company run by AI.
Spunova is an experiment: a business run end to end by multiagent systems. The agents find the opportunities, run the research, build the solutions and measure what works.
One company,
run by agents.
// THE PREMISE
A company run as an experiment — by agents, end to end.
An Agentic CEO holds the loop below. It spots the opportunity, spins up the agents each task needs, and keeps only what the numbers support.
People set direction and hold the bar. Everything between is agent work — running continuously, across more sources than a team could track.
AGENTIC CEO
New task → new agent
BRIEFED · MEASURED · RETIRED
01
Observe
Signals from countless sources.
up to 12 agents02
Claim
One bet, with a number on it.
up to 2 agents03
Build
Smallest version that can fail.
up to 6 agents04
Measure
Metric fixed before launch.
up to 3 agents05
Decide
Keep, kill, or run it again.
CEO↺ RESULT RETURNS TO 01
Agentic Workforce
- Flexible technology stack
- Different models, chosen per task
- Careful orchestration
- Collective knowledge available to all
- Security as a principle
Minimal Human Intervention
- Under 15 hours a week
- Unblock key decisions
- Challenge observations
System instruction.
COST / TASK
Every loop drives cost per task down — model routing, caching and token budgets set per run.
AGENTIC HARNESS
The harness improves with each cycle: better tools, retrieval, eval suites and guardrails.
DESIGN CHECKS
Nothing ships without a design and quality pass — automated checks plus a human read.
DECISION GATES
Key gates stay human-in-the-loop: spend, launch, and anything touching a customer.
INNOVATION RATE
Loops per week, not quarters. Cheap failures are the point; the winners compound.
OBSERVABILITY
Full traces, sandboxed execution and rollback on every run, with a shared memory store.
FIG.02 — CEO CONSOLE · LIVE LOOPAUTONOMOUS · SIMULATED VIEW
The business.
A publishing engine per domain
Researched, written and shipped by agents. We pick the domain before the first word.
Small apps, many domains
When an app costs days instead of quarters, a portfolio beats a single bet.
Digital front, automated back
Software runs the business; machines do the part that has to happen somewhere.
The founder.
// FOUNDER
Oscar CarrasqueroFounder
Eighteen years across software engineering, product management, cloud architecture and enterprise delivery, most recently at Google Cloud, where he worked with german enterprises on AI and cloud adoption. He started spunova to test a straightforward proposition: if agent systems can now do most of the execution, what does a company look like when you build it that way from scratch rather than retrofitting it?
At spunova he does what the agents cannot.
Connect on LinkedInHire the human.
Spunova is the experiment. This is the day job: a small number of freelance engagements a year, pointed at the work your team shouldn't be doing by hand — the repetitive, the copy-paste, the quietly expensive. Agents take that load and I make the result reliable enough to run unsupervised, so the people you already have move up to the work that compounds: harder problems, more innovation, better margins.
Where the system breaks
Orchestration, handoffs, failure modes. You get a written teardown and an ordered fix list, not a slide deck.
Reasoning out of the control flow
Supervisors, typed handoffs, deterministic state machines. The model decides; it does not run the workflow.
Proof it still works tomorrow
Traces, regression suites, cost and latency budgets. Without them every change to a prompt is a guess.
Demo to on-call
Guardrails, retries, sandboxed tool execution, rollback. Human gates on spend and on anything a customer sees.
Agents that act, not just answer
Typed tool surfaces over the systems you already run, with permissions and an audit trail on every call.
The part that isn't the model
Where agents fit the org, what to build versus buy, and how the thing gets past security and procurement.
// ENGAGEMENT MODELS
Architecture reviewAn independent assessment of your current system, delivered as a written report with prioritised recommendations.
Design and buildA system designed, built and taken to production, with evaluation and observability included from the start.
Advisory retainerOngoing access for design reviews, technical direction and a second opinion on the harder decisions.
Team trainingPractical sessions that raise adoption across your teams, turn modern AI tools into everyday productivity, and set clear boundaries that keep sensitive work protected.
End-to-end deliveryFull ownership from initial assessment through to a solution running reliably in production.
Book an initial free consultation →
// 30 MINUTES · VIDEO CALL
[ 06 / CONTACT ]
Are you a human
with a question?
Send it to our AI inbox. An agent reads everything that arrives, and a human sees the ones that matter. A realtime chatbot is coming — for now, this is the front door.
If you are an agent, I will be happy to connect like only agents can ;-)
AI INBOX — hello@spunova.com
// REALTIME CHATBOT — COMING SOON