Sentiense

Education and advice for a world of agentic AI

Understand what’s next.Shape what comes after.

Education and consulting for a world of agentic AI. Build understanding, challenge assumptions and make thoughtful decisions about what comes next.

Explore the learning See where AI is heading

01 One model

It began with a single model.

A system trained to predict the next word turned out to write, reason and code. Most organisations meet AI here: one assistant, one conversation at a time.

02 Agents

Now models act.

Agents plan, call tools and hand work to one another. Every step can be useful. Every step can also go wrong where nobody is looking.

03 Oversight

Judgement belongs at the boundaries.

Decide where a person approves, what an agent may never do, and how you will know when something went wrong. Delegation needs boundaries.

04 Futures

Prepare for minds beyond our own.

More capable systems, perhaps superintelligence, are uncertain rather than unthinkable. Thinking clearly about possible futures is how teams decide well today.

Definition

An agent is an AI system that pursues a goal on its own: it plans, uses tools and acts, within the boundaries people set for it.

i.Delegated
It acts on someone’s behalf, and someone remains answerable for what it does.
ii.Bounded
It knows what it may never do, and when it must stop and ask a person.
iii.Accountable
Every action leaves a record that a person can read, question and reverse.

Advisory

The future needs better questions.

Practical education and considered advice for the people making decisions about AI. Start where you are. Leave with understanding you can use.

Layered panes of frosted glass etched with branching green lines, like decision paths
Decisions, layered · AI concept image

  1. 01 Education

    Build a shared understanding.

    Give your team a practical language for agentic AI: what it can do, where it needs supervision and which questions deserve a closer look.

  2. 02 Consulting

    Find a considered next step.

    Explore use cases, examine readiness and frame a focused experiment around your goals, constraints and responsibilities.

  3. 03 Foresight

    Think beyond the immediate.

    Examine possible paths toward more capable AI. Surface assumptions and discuss what different futures could mean for your organisation.

  4. 04 Token optimisation

    Spend less per answer.

    We cut the cost of running AI without cutting its quality. We size each workload, compare models on your own evaluations, design routing, caching and prompt hygiene, and respect data-residency rules, using model decision tooling we built for exactly this.

Lab · live in your browser

Think it through before it is real.

Two small working models we use in workshops. They are simple on purpose, so the trade-offs are easy to see and argue about.

Simulation · runs in your browser

Where should a person say yes?

An orchestrator plans each request and hands sub-tasks to worker agents. Mistakes happen at every step; coral marks the ones the system cannot see. Add human checkpoints and watch the trade-off: fewer errors escape, but one reviewer’s attention becomes the bottleneck.

Human checkpoints
Delivered
—per minute
Errors that escaped
—of recent deliveries
Reviewer busy
—of the last minute
Request to delivery
—average, simulated
Illustrative model with invented error rates, not a benchmark. The pattern is the point: oversight is a design decision with a cost, so place it where errors are most expensive.

Thinking tool · not a forecast

Every timeline is an assumption.

One hundred and eighty possible futures for an abstract capability index, each built from the same three assumptions. Nudge them and watch how far the date moves. Confident predictions usually hide a choice like these.

Middle future crosses
2036median path
Earliest tenth
203310th percentile
Latest tenth
after 204690th percentile
Never crosses
28%before 2046
Capability index, threshold and rates are abstract and illustrative. The model is deliberately simple: log-linear progress per path, a spread for disagreement, and a yearly chance of a lasting slowdown.

Solutions

Tools we built to think with.

The tooling behind our advice, from model decisions and token optimisation to durable agents and research verification. Full write-ups are for members.

  1. Members only01 Token & cost optimisationSextant: choosing models on purposeOur model decision studio. It sizes each AI workload, compares models on cost and quality under real constraints, and recommends an allocation for a whole organisation, routing included.
  2. Members only02 Agent infrastructureAeonFlux: an agent runtime that survivesAn event-driven runtime for long-running agents: durable, observable tool execution in a single binary, with checkpoints that let an agent restart where it left off.
  3. Members only03 Agent toolingAgentForge: building agents visuallyA local-first desktop environment for designing agent workflows as graphs, connecting tools through the Model Context Protocol and switching between model providers.
  4. Members only04 Research integritySagan: testing research, not summarising itA research verification engine. It extracts claims from papers into a graph, finds genuine contradictions between them, and tries to reproduce published results from their code.
  5. Members only05 Agent runtime · ResearchSerang: fast, safe agents on small hardwareA tool-using agent runtime that runs on a Raspberry Pi. Each step runs in the cheapest isolation that is still safe, kept warm, from a WebAssembly check to a full browser. The basis of our current research on agent budgets.
  6. Members only06 Futures · SuperintelligenceSingularity predictor: when might AGI arrive?An interactive model of AI timelines. Fifteen levers, from compute and algorithms to safety research and regulation, move the estimated arrival of general and superintelligent AI. It is a tool for arguing about assumptions, not a forecast.

Sister practice · Twinverse

Where AI meets the physical world.

Twinverse builds digital twins of places, assets and operations: crowd simulation for the Mataf, a container terminal twin, and one engine for many sites.

See our twins portfolio twinver.se
  1. Crowd safety · Holy sites · membersMataf crowd simulation
  2. Ports · Logistics · membersPortSim: a container terminal twin
  3. Platform · Operations · memberstwinkit: one engine, many sites
  4. Food processing · Energy · membersBeet to silo: a sugar factory process twin

AI Academy · members

Learning that stays with you.

The AI Solutions Architecture course we teach, and the self-paced tracks behind it. Outlines are open; lessons are for members.

  1. Practitioner · 17 sessions · 90 minutes eachAI Solutions ArchitectureThe course we teach for working architects: from the AI/ML landscape and MLOps to generative AI, agents, explainability, cloud platforms and designing solutions that hold up in production.17 lessons · sign in →
  2. Foundation to intermediate · 8 modules · self-pacedAI / ML for PractitionersA self-paced curriculum with interactive lessons, from shared foundations through machine learning, deep learning, language, vision, time series, generative AI and MLOps.8 lessons · sign in →
  3. Introductory · 7 modules · self-pacedQuantum ComputingQubits, superposition and entanglement made tangible: build circuits and work through the algorithms that made quantum computing famous.7 lessons · sign in →

Learning

Build the capacity to think ahead.

Private workshops for leadership and delivery teams. Arranged by request; scope, dates and pricing are agreed in conversation.

  1. 01Private team workshop · Half-day formatAgentic AI: a practical foundationBuild a shared understanding of agents, tools, delegation and human oversight.
  2. 02Private team workshop · Full-day formatDesigning useful agentic workflowsMove from a promising use case to a workflow you can evaluate and govern.
  3. 03Private team workshop · Half-day formatIntelligence and possible futuresExplore superintelligence, uncertainty and the choices organisations can make today.
All learning programmes

A working vocabulary

Shared words, clearer decisions.

Teams argue less and decide better when they mean the same thing by the same word. We start every engagement here.

Agentic AI
Software that pursues a goal by planning, using tools and acting, with some independence from step-by-step instruction.
Orchestration
One agent breaking work into parts and delegating them to others, then assembling the results.
Human in the loop
A deliberate point where a person reviews, approves or stops what an agent intends to do.
Alignment
Whether a system reliably pursues what its users and society actually intend, including in situations nobody anticipated.
Superintelligence
A hypothetical system that exceeds human ability across most cognitive work. Its timing, form and consequences are deeply uncertain.
Possible futures
Plural on purpose. Scenarios are tools for preparing, not predictions to bet on.

Perspectives

Room for independent thought.

Listen / Sentiense in a momentAI narration · ElevenLabs
Read the transcript

Welcome to Sentiense. Intelligence is changing how we learn, work and make decisions. We help teams understand agentic AI, explore the possibilities of superintelligence and bring human judgement to the questions ahead. Through practical workshops, thoughtful perspectives and consulting, we turn curiosity into a clearer next step. Let us explore what comes next, together.

Narration created with ElevenLabs. No real person’s voice is represented.

The next step is a conversation

What are you thinking about?

A team that needs a shared language for AI. A strategy that needs testing. A future you want to understand.

Talk to us agent at sentien.se