A CV tells you where someone has been. It says far less about what they can deliver next.

In 2026 that gap matters more than ever. Companies need executives and technical specialists who can turn AI, data and modern engineering into measurable results, and who can work alongside AI agents rather than around them (Cornell CAHRS).

This article covers what has changed in tech hiring, which skills matter most, and a practical framework for hiring people and AI agents into the same team.

Why tech hiring has changed

AI adoption has moved on from experiments to redesigning how work gets done. Organisations are replacing fixed job descriptions with skills-based structures, deploying people according to what they can demonstrably do and how far they can grow.

Technology teams feel this first. Hiring is up across software development, technical management, IT operations, QA and entry-level technical roles, and fastest of all in AI (Linux Foundation). Supply hasn’t kept pace, so the competition for skilled people is sharpest in AI integration, cybersecurity and software engineering (Robert Half).

From job titles to capabilities

The old hiring question was: what role has this person held? The better question is: what can this person deliver, what evidence shows it, and how quickly will they learn the next thing?

A skills-based approach puts capabilities ahead of titles. It needs a clear skills map, visibility across teams, flexible deployment and continuous upskilling. It is fast becoming the thread that ties recruitment, onboarding, development, mobility and retention together (Randstad Digital).

For technical and executive hires, look for evidence that someone has:

• Shipped products or systems into production
• Led delivery across functions
• Made sound trade-offs under uncertainty
• Turned business goals into technical execution
• Checked AI output critically instead of accepting it as given
• Built or led blended teams of employees, contractors and AI agents

This is why critical thinking tops the wish list. Korn Ferry found that 73% of talent leaders name critical thinking and problem-solving as the skill they most need in 2026 (Korn Ferry). AI can produce the output. People still have to judge its quality, risk, context and consequences.

How teams are being reorganised

AI is changing how work is organised as well as who does it. Instead of rigid job families, organisations are redesigning roles around tasks and deciding, task by task, whether a person or an AI system should do it (Aon). Four team shapes are becoming common:

• Product squads bring engineering, design, data, commercial and AI skills together around one outcome.
• Platform teams build the shared infrastructure, data pipelines, security controls and agent capabilities everyone else uses.
• AI enablement teams govern model use, prompt standards, evaluation, risk and adoption.
• Hybrid delivery pods mix employees, contractors, partners and AI agents to deliver at business speed.

The result is flatter and more networked. Performance reviews, rewards and governance need to follow, recognising skills, outcomes and adaptability rather than tenure or a fixed role definition (AIHR).

The skills in highest demand

Demand is spreading well beyond traditional software engineering. In 2025, job postings for AI, machine learning and data science more than doubled, and so did postings for security roles. Cybersecurity engineers alone accounted for 20,000 new postings (Robert Half).

Leaders, meanwhile, need to be fluent in both commercial strategy and technology delivery. The most effective executives can connect AI spending to revenue, margin, risk, customer experience and operating capacity.

AI is changing recruitment itself

Talent acquisition is going AI-first. Sourcing, screening, scheduling, candidate messages, CV rediscovery, background checks and onboarding logistics are increasingly automated (SHRM).

Buying more tools doesn’t fix hiring on its own, though. Most organisations have increased their spending, yet few can grow hiring volume without growing the recruiting team.

The model that works is human-in-the-loop AI. Automate the repetitive work, and keep human judgement for assessing candidates, aligning stakeholders, judging team fit, making senior appointments and protecting fairness.

Hiring people and AI agents into the same team

This is the thinking behind ActivoNEO: hire for both people and AI agents, and replace titles and CVs with capability backed by evidence. For people, that means skills verified through tests, work samples, portfolios and simulations. For AI agents, it means benchmarking task success, tool use, reliability, cost per task and guardrails.

Seven steps to build an AI-enabled team:

1. Define the outcome first. Start with the business result, not the job title.
2. Break the work into tasks. Decide which belong to people, which to AI agents and which to a hybrid workflow.
3. Specify the evidence. Ask for portfolios, simulations, technical assessments or delivery case studies.
4. Benchmark the agents. Measure accuracy, reliability, cost, speed and how they fail.
5. Design the operating model. Agree decision rights, escalation paths and where humans oversee the work.
6. Hire for adaptability. Prioritise learning speed, judgement and the ability to work across functions.
7. Measure by outcomes. Track delivery speed, quality, adoption, cost and business impact.

The hires that unlock growth

For a growing company, the highest-return hires are rarely the biggest teams. They are the few people who unlock execution:

• A Chief AI or Data Officer who connects AI strategy to commercial value and governance.
• An AI engineer or agent-orchestration lead who gets reliable systems into production.
• A product leader who turns customer problems into prioritised, measurable delivery.
• A data engineering lead who builds data foundations people trust and use.
• A security and risk leader who enables innovation without exposing the business.
• A talent strategist who builds internal mobility, reskilling and a workforce model for people and agents.

The bottom line

Tech hiring is moving from volume to precision: fewer unvetted CVs and more validated capability, fewer rigid roles and more adaptable skills, less manual admin and more human judgement where it counts.

The organisations that win won’t simply hire more engineers. They will build an operating system for talent: one that identifies the capabilities they need, verifies them with evidence, puts people and AI agents where each works best, and keeps developing the workforce as the technology moves (Cornell CAHRS).

Where Activo can help

Whether you are hiring for your next stage or looking for your next role, these are good places to start.

Sources: Cornell CAHRS; Linux Foundation Tech Talent Report 2026; Robert Half; Randstad Digital; Korn Ferry; Aon; AIHR; SHRM; Yahoo Finance.

The post Tech Talent Acquisition in 2026: Hiring for Capability, Execution and Human–AI Teamwork first appeared on Activo Group.

For years, supply chain teams were judged on one question: how efficient is the plan? Lowest cost, leanest inventory, tightest routes.

That question still matters. In 2026, though, a second one matters more: how quickly can the plan change when the world does?

AI is what makes that second question answerable. It lets a supply chain sense demand shifts earlier, rehearse disruptions before they happen and act on them in hours rather than weeks. This article looks at the technology behind that shift and the five design choices that turn an efficient supply chain into an adaptable one.

Why the old model is under pressure

Traditional supply chains were built around accurate forecasts, long planning cycles and cost minimisation. That works when demand is steady and supply is predictable. It struggles with what most businesses now face: geopolitical shocks, shifting trade policy, inflation, labour shortages and disruptions that recur rather than arrive once a decade (Yahoo Finance).

So companies are modernising around end-to-end visibility. The aim is to put procurement, production, inventory, logistics and customer demand into one decision environment, instead of letting each function optimise for itself and pass the problems to the next.

Optimisation and adaptability solve different problems

They aren’t rivals. They answer different questions, and each fails in its own way if you push it too far.

The practical answer is adaptive optimisation. Use AI to find the most efficient plan, then design the network so that plan can be rewritten quickly when its assumptions break. That is why supply chain leaders now treat resilience and governance as part of the foundations, not an optional layer on top.

Five technologies changing how supply chains work

1. Demand sensing that changes decisions

AI forecasting models combine sales history with promotions, weather, market signals and other external indicators. They pick up changes earlier and say more precisely where stock should sit, when to replenish and which products need extra buffer (ASCM).

The test is simple: does the forecast change what happens next? If it feeds replenishment, production, procurement, carrier choice and warehouse staffing, it earns its keep. If it ends up as another dashboard, it doesn’t (Economic Times).

2. Control towers that coordinate, not just monitor

Control towers started as dashboards that showed where things were. They are becoming orchestration hubs that combine live operational data with predictions, so one team can coordinate suppliers, carriers, warehouses and customers from the same view. A useful control tower answers four questions:

• Which orders are at risk in the next 48 hours?
• Which suppliers, lanes or sites are showing early signs of stress?
• What is the best response if a port, plant or carrier fails?
• Which trade-off gives the best mix of service, cost and cash?

3. Digital twins for rehearsing disruption

A digital twin is a virtual model of a factory, warehouse, transport network or product flow. It lets you test “what if” before you commit money or capacity (Microsoft).

Unilever shows what this looks like at scale. Its AI-enabled twins monitor equipment and production lines, predict maintenance needs and make sure spare parts are on hand. At one plant it reported 20% less waste and 10% more capacity, with fewer stoppages and defects at others (Yahoo Finance).

4. Agentic AI: from insight to action

This is the biggest shift. Conventional automation follows fixed rules. Agentic systems are given a goal and can reason, plan, decide and act towards it with limited human input (DHL).

In logistics, that means an agent can reroute a shipment around a closed port, add transport capacity, flag a coming stock shortage or start a replenishment order. DHL ranks agentic AI among the most impactful logistics trends because it moves supply chains from watching problems to dealing with them.

Agents are leaving the pilot stage quickly. That means questions about data quality, guardrails and oversight need answers now, not after the first agent is already making decisions.

5. Warehouse automation and physical AI

AI is reshaping the physical layer too. Warehouse orchestration assigns work to robots based on priority, stock position and equipment status, while predictive maintenance cuts unplanned downtime (WNS).

The next step is “physical AI”: robots and machines that use AI to work better in warehouses, distribution centres and transport. Microsoft describes it as combining 3D simulation, discrete-event simulation and digital twins to predict outcomes, tune performance and prescribe the next action.

How to build a network that changes efficiently

The most valuable supply chains aren’t necessarily the leanest. They are modular, visible and ready to decide. Five design choices get you there.

1. Build visibility before autonomy

An AI agent can’t reroute a shipment, rebalance stock or replace a supplier if it can’t see the data. Connect procurement, inventory, transport, supplier performance and demand signals first. Fragmented systems make forecasts unreliable and responses slow.

2. Plan with scenarios, not a single forecast

Instead of planning around one expected outcome, run scenarios: a demand spike, a supplier failure, port congestion, a tariff change, a carrier running out of capacity. AI can compare the responses and put a number on the service, cost and inventory impact of each before any of it happens.

3. Tier decisions by risk

Not every decision needs a person to sign it off. Low-risk, easily reversed actions can be automated. Decisions with a big or lasting impact should keep a human in the loop.

4. Design options into the network

Adaptable networks have more than one supplier for critical items, flexible transport modes, regional inventory and fulfilment paths that can be reconfigured. The goal isn’t to remove every inefficiency. It’s to avoid a design where one failure stops everything.

5. Measure adaptability

Cost per order, inventory turns and on-time delivery still matter. Add measures that show how well the network copes with change:

• Time to act: hours from a disruption being detected to a corrective action being taken
• Data freshness: share of decisions made on current, integrated data
• Shock accuracy: forecast error during demand shocks, not just in normal weeks
• Concentration risk: dependence on single suppliers and carriers
• Scenario coverage: share of critical products and lanes with a tested response plan
• Promise accuracy: how often customer delivery promises hold during disruption

What comes next

Over the next few years, supply chains will become more predictive, more autonomous and more elastic. AI will take on routine operational decisions, and people will spend their time on exceptions, strategy, supplier relationships, governance and ethical oversight (DHL).

The strongest supply chains will combine three capabilities:

1. Optimisation to remove waste and improve unit economics.
2. Intelligence to spot change earlier and put numbers on the trade-offs.
3. Adaptability to reconfigure sourcing, inventory, production and delivery quickly.

Treat AI as a better forecasting tool and you will get some efficiency gains. Use it to build a connected network that runs scenarios and acts on them, and you will be far better placed to protect service, margins and customer trust when the next disruption arrives.

Where Activo can help

If this sounds like your operation, these are the places we usually start.

Sources: Yahoo Finance, supply chain modernisation; Yahoo Finance, Unilever; ASCM Top 10 Trends 2026; DHL Logistics Trend Radar; Microsoft; WNS; Economic Times.

The post Supply Chain Optimisation in the Age of AI: Why Adaptability Is the New Efficiency first appeared on Activo Group.

Founders can measure hundreds of things: traffic, clicks, MRR, daily users, feature usage, NPS, pipeline. Early-stage momentum usually comes down to five: retention, growth rate, customer acquisition cost (CAC), conversion and lifetime value (LTV), plus one that investors check first, payback period. This guide explains why each one matters, how to measure it, what good looks like today, which secondary metrics to watch, and how to turn them into a dashboard and a daily alert routine. Updated October 2026.

The five core metrics at a glance

A startup is a machine that attracts users (growth and CAC), converts them (conversion), keeps them (retention) and earns more from them than they cost (LTV and payback). If one part breaks, growth gets expensive, churn becomes a leak and fundraising gets harder.

1. Retention: the strongest signal of a big business

Why it matters: retention is the only metric that proves value rather than marketing. Every other number depends on it: growth stalls if customers leave as fast as they arrive, CAC never pays back if customers churn before month 12, and LTV is mostly a function of how long people stay. It also compounds the other way. With 101% median net revenue retention, the typical SaaS company has to replace almost all of its growth with new customers every year, while companies above 120% grow even if sales stop.

How to measure it: group users by the week or month they started (a cohort) and track the share still active after 1, 4 and 12 weeks. For paid products, also track logo churn (customers lost) and revenue churn (money lost). Define active by the frequency your product should have: daily for habits, weekly for workflows, monthly for billing or finance tools.

Pitfall: celebrating signups while retention is flat. Fix the leak before you pour in more water.

2. Growth rate: are you compounding fast enough?

Why it matters: growth is what turns a good product into a venture-scale company, and small differences compound into huge gaps. Paul Graham’s rule of thumb for Y Combinator companies is 5–7% a week, which is roughly 24–34% a month. The best growth comes from channels that do not need more spend: word of mouth, search, partnerships, earned media and product loops where users invite users. Even a viral factor (invites per user × invite conversion) of 0.1–0.2 lowers CAC over time.

Pitfall: measuring growth in a vanity number. Pick one north-star metric that reflects value delivered, such as MRR, weekly active teams, bookings or gross merchandise value, and report its growth every week.

3. CAC: what does a paying customer really cost?

Why it matters: CAC decides whether growth creates or destroys value. If it costs £3,000 to win a customer who brings £1,500 of gross profit before churning, every sale makes you poorer. At the median, B2B SaaS companies now spend $2.00 of sales and marketing to win $1.00 of new annual revenue, so efficiency is what separates fundable companies from the rest.

How to measure it: CAC = all sales and marketing cost / new paying customers. Use fully loaded CAC: ad spend, discounts, sales salaries and commissions, onboarding and support time, tools and agencies. Track paid CAC (paid channels only) and blended CAC (all customers), and break both down by channel.

Pitfall: “Our CAC is £10 because that is what we spend on ads.” If winning the customer also took a sales call, a discount and an onboarding session, your real CAC is several times higher.

4. Conversion: do users reach the aha moment and pay?

Why it matters: conversion is the cheapest growth lever you have. Doubling activation from 20% to 40% doubles the customers you get from the same traffic and halves your effective CAC, without spending more. It also shows where the product, not the marketing, is failing.

How to measure it: track each step separately: visit to signup, signup to activated (reaching the aha moment), activated to paying, and paying to retained. For marketplaces, conversion is liquidity, and you need both sides: the share of listings that sell (supply) and the share of buyers who find something and transact (demand).

Pitfall: measuring only one side of a marketplace, or only the final purchase. Most leaks happen between signup and first value.

5. LTV: how much profit does a customer bring?

Why it matters: LTV sets the ceiling on what you can afford to spend to win a customer. It turns retention, pricing and margin into one number, and paired with CAC it shows whether the business model works at all.

How to measure it: for subscriptions, LTV = monthly revenue per customer × gross margin / monthly churn rate. For transactions, LTV = average order value × take rate × gross margin × purchases per year × years retained. Always use gross profit, not revenue: a £100 customer at 30% margin is worth less than a £60 customer at 80%.

LTV to CAC: divide LTV by CAC. Around 3× is the usual benchmark for a healthy model, while above 5× often means you are under-investing in growth. Be wary of LTV figures built on a few months of data and an assumed five-year life.

Bonus: payback period, the metric cash cares about

Why it matters: payback = CAC / monthly gross profit per customer. It tells you how long your cash is tied up in each new customer, which matters more than LTV when money is limited. The 2025 median was 16 months, with 11 months for products priced under $5k a year and 22 months for enterprise deals. Top-quartile companies pay back in six months or less, so they can recycle cash into growth twice a year.

Secondary metrics worth monitoring

Once the five core metrics are in place, add these to explain why the core numbers move. You do not need all of them: pick the ones that match your model and stage.

What an ideal startup dashboard looks like

A good dashboard answers three questions in under a minute: are we on track, what changed, and what should we do about it? Keep it to one screen, put the north-star metric at the top, show every number against a target and a trend, and use the same definitions everywhere so the team never argues about what a number means.

Design rules: no more than about 12 numbers on the main view, every metric has an owner, targets are written down, and anything that has not changed is hidden. A dashboard nobody acts on is decoration.

Start each day with actionable alerts

Dashboards still depend on someone remembering to look. The better routine is the reverse: the numbers come to you, and only when something needs a decision. That is how the 360 Orchestrator in ActivoCEO works. It monitors your KPIs, email, commitments and risks across the systems you already use, prioritises what needs attention today and delivers a five-minute daily briefing with an exception-based dashboard. Low-risk, reversible tasks run automatically, while consequential actions wait for your approval.

When a metric is weak: where to start

Your 30-day metrics plan

1. Days 1–3: Define your north star and the core five. Write one-line definitions for each, including what counts as an active user and a new customer, and agree them with the team.
2. Days 4–10: Connect the data. Pull billing, product analytics, CRM and ad spend into one place. Calculate fully loaded CAC and gross margin from real costs, not estimates.
3. Days 11–14: Build cohort views. Create monthly cohorts for retention and revenue, and compare your latest three cohorts with older ones.
4. Days 15–20: Set targets and owners. Give each metric a target for the next quarter and a named owner. Add two or three secondary metrics that explain the weakest core number.
5. Days 21–25: Build the one-page dashboard. North star on top, core five against targets, drivers and cash below. Hide anything nobody will act on.
6. Days 26–30: Switch on daily alerts. Set thresholds for the triggers above so the exceptions come to you each morning, and review the dashboard with the team every week.

Sources: Benchmarkit, 2025 B2B SaaS Performance Metrics Benchmarks · Aleph and Benchmarkit, CAC payback benchmarks 2026 · Paul Graham, Startup = Growth. Dashboard and briefing examples are illustrative.

The post 5 Core Startup Metrics That Actually Matter first appeared on Activo Group.

Consciousness did not switch on all at once. Across billions of years, living things gained capabilities in layers: sensing and acting, then integrating signals, remembering, predicting, modelling themselves and finally noticing their own uncertainty. AI is passing the same checkpoints, in both software (digital AI) and robots (physical AI), but on a clock that runs millions of times faster. This guide maps the two paths, shows where AI stands at each step today, what comes next, and how to use the map to build more capable products in your own startup. Updated October 2026.

Two things to keep separate: capability and experience

Science still has no agreed definition or test for consciousness. What we can track is the set of capabilities that evolution kept selecting because they helped organisms survive in messy, changing environments. Those same capabilities are exactly what autonomy in AI requires. This article is about capability. Whether an AI system could ever have experience is a separate and open question. A 2023 review by 19 researchers led by Patrick Butlin and Robert Long derived 14 indicator properties from leading theories and concluded that no current AI system is a strong candidate for consciousness, but also that there are no obvious technical barriers to building systems that meet more of the indicators.

The six checkpoints, in nature and in AI today

Each checkpoint below solved a real survival problem. For each one, the table shows where digital and physical AI stand in 2026 and what the next step looks like.

Why digital AI evolves faster than physical AI

In biology, generation time sets the pace of evolution. Bacteria can adapt in days, elephants take millennia. AI has the same split. A software agent can be retrained, tested against millions of tasks and redeployed in weeks. A robot has to be built, shipped, made safe and taught in the physical world, where every training example costs time and money. That is why the task length digital AI can handle alone has doubled every few months, while the number of industrial robots installed each year took a decade to double.

Physical AI is closing the gap by borrowing the digital playbook. World models create synthetic environments so robots can practise millions of times before touching a real object. Fleets pool their experience, so every vehicle learns from every other. Amazon, now running over 1 million robots across more than 300 facilities, says its DeepFleet AI model improved robot travel efficiency by 10%.

Prediction is the engine of both minds and agents

A leading view in neuroscience, known as predictive processing, holds that brains do not passively record the world. They predict what the senses will report and learn from the gap. The free energy principle and active inference, associated with Karl Friston, go further: organisms act to make their predictions come true. Anil Seth describes perception as a controlled hallucination, kept honest by sensory data.

AI has followed the same path. The step from chatbots that complete text to agents that complete jobs came from adding prediction and checking: reasoning models that plan before they answer, and world models that simulate what will happen if they act. Prediction turns intelligence into a simulation engine that can ask “what happens if I do this?” before paying the cost of doing it.

Three theories of consciousness, and what each teaches AI builders

No theory has won. In 2025 the Cogitate Consortium published in Nature the results of an adversarial collaboration, where rival camps agreed predictions in advance and independent labs tested them on hundreds of participants. Neither Global Workspace nor Integrated Information Theory came through cleanly. Each theory still offers a useful design lesson.

Different brains, same abilities: convergent evolution

Evolution found intelligence more than once. An octopus has about 500 million neurons, roughly two-thirds of them in its arms, and solves puzzles with a brain plan unlike ours. Ravens pack around 2 billion neurons into a walnut-sized brain and plan for future needs. Humans run 86 billion neurons on about 20 watts. The Cambridge Declaration on Consciousness (2012) stated that many non-human animals, including birds and octopuses, have the neurological substrates for conscious experience. The New York Declaration on Animal Consciousness (2024) went further, finding a realistic possibility of experience in all vertebrates and many invertebrates.

The lesson for AI: there is no single magic architecture. Transformers, diffusion models, world models and robot control policies are different implementations converging on the same functions. Bet on the capability your customer needs, not on one model or vendor.

How AGI will keep evolving: six shifts to watch

If the history of minds is a guide, the next gains come from deepening the later checkpoints rather than from size alone. Here is what each shift looks like in digital and physical AI, and the signal that it is arriving.

Waymo shows what mature self-monitoring looks like in practice. Its vehicles start every request for remote help themselves, receive advice rather than remote control, and about 70 remote assistance agents support a fleet of around 3,000 vehicles. That is roughly one human per 40 robots, and it is the right model for digital agents too: autonomy by default, help on request, humans accountable.

Use evolution’s checklist to grade your own AI product

Score your agent or AI feature from 0 to 2 on each checkpoint. Most early products score 4 to 6 out of 12. The lowest score is your next build priority.

Examples: applying the checklist to your startup

A 30-day plan to move your product up the checklist

1. Days 1–5: Score and measure your time horizon. Grade your product on the six checkpoints. Then measure how long it runs on a real task before a human must step in. That is your own time horizon, and the number to grow.
2. Days 6–12: Fix sensing and integration. Connect the two or three data sources and actions the job truly needs, and give every step one shared task state instead of separate prompts.
3. Days 13–18: Add memory that changes behaviour. Turn corrections and outcomes into rules the agent applies next time, and review them weekly.
4. Days 19–24: Add prediction and checking. Make the agent write a plan, dry-run risky actions and verify results before it reports done.
5. Days 25–30: Add calibrated escalation. Define when to ask for help by confidence and stakes, log every escalation, and re-measure your time horizon and error rate against day 1.

Closing thought

The story of consciousness is a story of adaptive control: better models, better integration, better self-monitoring, again and again, because the world keeps changing. The AGI race is the same story told in months rather than millions of years. AGI is unlikely to arrive as one breakthrough. It will look like systems, digital and physical, that stay coherent while the world changes, know when they are unsure and keep improving from experience. Startups that build those capabilities into their products now will be ready for each new model rather than replaced by it.

Sources: METR, Measuring AI ability to complete long tasks · METR time horizon results · Epoch AI, trends · IFR, World Robotics 2025 · IFR humanoid count, via AI Weekly · Amazon, 1 million robots and DeepFleet · Waymo 500,000 weekly rides · Waymo, Advice not control · The Decoder on Genie 3 · Max Planck Institute on the Cogitate study · Butlin, Long and others, Consciousness in AI · NBC News on the New York Declaration.

The post How Consciousness Evolved: The AGI Race Follows the Same Path first appeared on Activo Group.

Generative AI gave businesses content and copilots. Agentic AI gives them software that plans, decides, transacts and fixes its own errors, and increasingly does so by talking to other agents. When the users of a platform include agents that can search, negotiate, book and pay at machine speed, network effects change too. This guide explains how, shows where the new advantages form, and gives you concrete ways to build them into your startup. Updated October 2026.

Why this matters now: lots of pilots, little profit

Most companies have tried AI, few have made money from it. McKinsey’s November 2025 survey found 62% of organisations experimenting with agents but only 23% scaling one, and just 39% reporting any profit impact from AI. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027 because of rising costs, unclear value or weak risk controls, and estimates only about 130 of the thousands of agentic vendors are genuine. The gap is the opportunity: agents that complete a whole process, not just assist with one step, are what turn usage into measurable results.

From copilot to agent: what actually changes

A copilot waits for a prompt and answers. An agent is given a goal, breaks it into steps, calls tools, checks rules and handles exceptions until the job is done. Think of a bright intern versus a capable operator. The real gains come when you redesign the workflow around the agent, running steps in parallel and routing exceptions automatically, rather than dropping it into an old sequential process.

Why teams of specialist agents beat one super-agent

The winning design is not one agent that does everything, but several narrow agents that check each other: one analyses, one executes, one monitors, one enforces policy and one escalates to a human. Narrow agents are easier to test and govern, they work in parallel, and control becomes an explicit layer rather than a hope.

Startup example: an accounts-payable product could run an extraction agent that reads invoices, a matching agent that checks them against purchase orders, a policy agent that flags anything over the approval limit, a payment agent that schedules payment, and an audit agent that logs every decision. Each one can be measured and improved on its own, and the customer sees one finished outcome: invoices paid correctly and on time.

The plumbing: protocols that let agents work together

Agents only create network effects if they can connect. Without shared standards you get protocol islands, with agents trapped inside one vendor. In 2025 and 2026 the industry converged on a small set of open standards, most now under neutral foundations.

Why this matters for network effects: open standards create gravity. Every new MCP server makes every assistant more useful, and every new assistant makes publishing an MCP server more valuable. The products that plug in early get used by more agents, collect more outcomes and become the default choice.

Five network effects of the agentic economy

How to apply this to your startup: four worked examples

Run it like a network: from uptime to completion

Agent workflows hop across identity, payments, stock, messaging and CRM in one chain. A server that is up 99.9% of the time is no comfort if one step in the chain fails silently. The question changes from “is it up?” to “did the whole job complete?”, and data quality becomes existential: bad data no longer produces a wrong chart, it produces a wrong payment.

The agent-ready scorecard

Your 60-day plan to become agent-ready

1. Days 1–7: Pick one workflow with clear money attached. Choose a process where cost, cycle time or errors are measurable, and define “done” as the completed outcome, not a conversation.
2. Days 8–15: Map the chain. List every system, decision and hand-off in the workflow. Mark which steps an agent can do alone and which need human approval.
3. Days 16–30: Build specialist agents with guardrails. Split the work into narrow agents, add a policy agent and an audit log, and test against 50 real cases.
4. Days 31–40: Open the door. Publish an MCP server for your three most valuable actions, and add an A2A endpoint if other agents should be able to hire yours.
5. Days 41–50: Instrument completion. Track end-to-end completion, overrides, exceptions and data freshness on one dashboard, and publish a reliability page.
6. Days 51–60: Seed the network. Get listed in MCP and agent directories, sign two integration partners, and price at least one offer per completed task.

When agents can find, hire and pay each other, defensibility shifts from who has the most users to who is easiest to connect to, most trusted and best at finishing the job. Start with one workflow, make it agent-ready and let the network do the compounding.

Sources: Linux Foundation, Agentic AI Foundation launch · Linux Foundation, A2A one year on · Digital Commerce 360 on Gartner B2B forecast · Outlook Business on Gartner agentic predictions · McKinsey, The state of AI in 2025 · PaymentsJournal on OpenAI and Stripe.

The post Agentic AI and the Next Wave of Network Effects: When AI Agents Talk to Each Other first appeared on Activo Group.

Network effects create outsized growth, but they come with a built-in pricing problem. The thing that makes a network valuable, lots of active users, is exactly what you lack at the start, so you are pushed to price at zero. Charge too early and you never reach critical mass. Never charge and the value you create leaks to users, partners or competitors. This guide shows how to light the network, how to tell when it has tipped, and which pricing models capture value without breaking it, with examples you can apply to your own startup. Updated October 2026.

The paradox in plain terms

A network is weak when it is small and strong when it is big. Early on, the product may be close to useless without other users, so asking for money is a non-starter. Later, the network is worth far more than any single feature, and not charging means the value exists but your business does not capture it. The answer is not choosing growth or revenue. It is sequencing: subsidise to ignite, prove the tipping point, then charge where value is clearest and price sensitivity is lowest.

Beat the cold start: five ignition strategies

Before network effects work for you, they work against you. A new network with no users has no network value, and every visitor asks “who else is here?”. Your goal is the tipping point, where the value of joining finally exceeds the effort and cost of joining.

Find your hard side, then price around it

Two-sided networks are never symmetrical. One side is harder to acquire, more sensitive to economics and decides whether the other side has anything to find. Subsidise that side, and charge the side with clear, measurable return on what it pays.

When to start charging: tipping-point signals

Monetisation timing is about network maturity, not revenue targets. Start shifting from subsidy to value capture when most of these signals are true, and change one side at a time.

Pricing models that capture value without killing the network

Match pricing to network maturity

The most common mistake is running established-stage pricing on an embryonic network. The second is staying free so long that users learn the product is worth nothing.

Worked examples for your startup

Quick answers founders ask

Your 90-day network pricing plan

1. Days 1–10: Define your atomic network. Choose the smallest segment where the network can work, and set a density target, such as 40% of salons in one district.
2. Days 11–20: Identify the hard side. Interview ten people on each side. The side that is harder to win and more price-sensitive is the one you subsidise.
3. Days 21–45: Ignite by hand. Seed supply, match the first 100 transactions yourself and add single-player value so early users stay even when the network is thin.
4. Days 46–60: Instrument the tipping point. Track liquidity, organic share of new users, cohort retention and value per user on each side every week.
5. Days 61–75: Test pricing on one side. Introduce a fee or paid tier for one cohort of the side with clear return, and watch liquidity on the other side.
6. Days 76–90: Roll out with protection. Extend pricing gradually, grandfather early adopters and keep a free or threshold tier so the network keeps growing.

The winners do not choose between growth and monetisation. They sequence them: ignite with subsidies, prove the network has tipped, then capture value with precision.

Sources: NFX, 70% of value in tech is driven by network effects · CNBC on Shopify app store fees · Apple App Store Small Business Program · Etsy fees and payments policy · Andrew Chen, The Cold Start Problem.

The post Network Pricing Paradox: How to Ignite Explosive Network-Effect Growth and Capture Value first appeared on Activo Group.

Most founders talk about viral growth. Far fewer design for network effects, and that is where the durable value sits. Virality brings users in. A network effect is the reason they stay, because each new user makes the product more valuable for everyone already there. This guide covers the maths behind that, which types of network effect matter for AI products, and a 30-day plan to test whether you really have one.

Virality grows you. Network effects keep you.

A simple test: if growth stopped tomorrow, would the product get worse for the people already using it? If not, you have virality without defensibility. You are in the “fresh produce” business, constantly shipping new content or features to give people a reason to come back.

The four defensibilities left in a digital market

Digital distribution has eroded the old moats of geography, shelf space and capital. Four remain: scale (cost advantages at volume), embedding (hard to rip out of workflows), brand (trust and preference) and network effects. In the NFX data, the companies built on embedding, scale and brand typically topped out at $1–2 billion. The outliers were network businesses, because their value can grow faster than their costs.

The network maths, drawn

Four models describe how value scales with the number of users (n). Treat them as lenses, not laws.

The founder insight: do not put Metcalfe or Reed into a financial model. Plan with something closer to n log n and let the upside surprise you. What the curves do tell you is where to design: products where users form groups (Reed) or direct links (Metcalfe) have far more headroom than products where users only consume (Sarnoff).

The network effects worth designing for

“We have a data moat”: three things founders confuse

Embedding is robust but it is not a network effect. The strongest AI defensibility combines a real data network effect with deep workflow embedding. To check for the real thing, use the Waze test: is data captured automatically as people use the product, does it improve outcomes quickly, does value keep growing rather than plateau, is the data central to the product, and can users feel that collective usage makes it better?

How to start a network from zero

• Single-player mode. Be useful to the very first user on their own. Tool first, network second.
• Subsidise the hard side. In marketplaces, supply is usually harder to win than demand. Secure it with credits, guarantees or done-for-you onboarding.
• Use the technology window. Generative AI is resetting workflows. Use the reset to deliver a step change in speed or outcomes that pulls people onto a new network.
• Move beyond fresh produce. If your agent generates content, shift towards user-to-user interaction, shared data loops, templates and integrations that others build on.

Your 30-day network effect plan

1. Week 1: Score yourself. Answer the seven questions below with your team. Two or fewer yeses means you have growth tactics, not defensibility.
2. Week 2: Map your nodes. List who joins (users, suppliers, developers, data sources), what each gives the others, and which side is hardest to win.
3. Week 3: Design one loop. Pick a single mechanism, such as single-player value, a hard-side subsidy or a shared data loop, and ship the smallest version.
4. Week 4: Measure density, not just growth. Track the share of users with at least one active connection, fill rate or time-to-match, and whether retention rises as networks get denser.

The seven-question self-test

1. Do users stay because other users are there?
2. Does value rise for existing users when new users join?
3. Does leaving mean losing connections, liquidity or collective intelligence?
4. Does usage create data that improves outcomes for everyone, quickly?
5. Does value keep growing past “good enough”, or does it plateau?
6. Are we embedded in workflows so removal is painful?
7. Can others build on top of us while we sleep?

Virality can ignite growth. Network effects, paired with embedding and data loops, are what make it last. For agentic products the goal is not to generate more. It is to design interactions so the system gets more valuable as the network grows.

Sources: NFX, 70% of value in tech is driven by network effects; NFX, The Network Effects Manual; Briscoe, Odlyzko and Tilly, Metcalfe’s Law is Wrong (IEEE Spectrum, 2006).

The post Network Effects for Startups: The Math of Defensibility first appeared on Activo Group.

Network effects are usually explained with apps: the more people join, the more useful the product becomes. The same maths runs your life. Who you know decides what you hear about, which doors you notice, what feels normal and even how long you live. This guide pulls together the research on friendship, careers and culture, shows how to read your own network like a founder reads a growth chart, and gives you a 30-day plan to redesign it. Updated October 2026.

What a network effect means in your life

A network effect happens when each new connection changes the value of the whole network. In personal life, value comes in four forms. Each one is backed by solid research, and each one can be designed rather than left to chance.

Your brain has network limits: the Dunbar layers

Anthropologist Robin Dunbar found that human relationships fall into layers that grow roughly threefold each step. By his estimate we give about 40% of our social time to the closest five people and about 60% to the closest fifteen. Time is the constraint, so the question is not how many people you know, but how you spread your hours.

Dinner party maths: the number of possible two-way relationships in a group is n(n − 1) / 2. Six people make 15 pairs, ten make 45, fifteen make 105. That is why friend groups splinter as they grow and why good hosts keep tables small or give the evening a structure.

How friendships actually form: it takes hours

In a 2018 study, communication researcher Jeffrey Hall followed adults who had just moved and students starting university. Going from acquaintance to casual friend took about 50 hours together, to friend about 90 hours, and to close friend more than 200 hours. Only social time counted fully. Hours spent working side by side counted for much less unless people also talked and relaxed together.

Five conditions make those hours easy to accumulate. Change these inputs and friendships follow.

Careers run on weak ties

In 1973 sociologist Mark Granovetter asked people who had recently changed jobs how they heard about the role. Of those who found it through a personal contact, most had seen that contact only occasionally or rarely. Close friends know what you already know. Acquaintances connect you to other worlds.

Half a century later, a team from MIT, Stanford, Harvard and LinkedIn tested the idea causally. Experiments across 20 million LinkedIn users over five years, which created about 2 billion new ties and 600,000 job changes, found that moderately weak ties helped job mobility most. The very weakest ties were too thin to help, and strong ties added little new information.

How to use it: keep a list of 30 to 50 moderately weak ties, such as former colleagues, people from courses and friends of friends. Reconnect with two a week with something useful: an article, an intro, congratulations on a move. When you need help, the channel is already warm.

Behaviour spreads: three degrees of influence

Nicholas Christakis and James Fowler used decades of data from the Framingham Heart Study to trace how behaviours move through friendship networks. Happiness, weight gain and quitting smoking all spread, and the effect reached friends of friends of friends.

Culture works the same way at scale. In a 2018 experiment published in Science, Damon Centola and colleagues showed that once a committed minority reached about 25% of a group, it could flip the group’s established convention. Norms feel fixed until a quarter of the room changes its mind.

Networks shape opportunity across society

The largest study of its kind, using 21 billion Facebook friendships, found that economic connectedness, meaning friendships between people of lower and higher income, was the strongest predictor of upward mobility that researchers measured. Low-income children who grew up in places with plenty of cross-class friendships earned about 20% more as adults.

Technology also changes how networks form. Research led by Stanford’s Michael Rosenfeld found that meeting online became the most common way heterosexual couples in the US met, at 39% in 2017, while meeting through friends fell from 33% in 1995 to 20%. Remote work, online communities and AI tools are a similar window today. They make it easier to reach new networks, and easier to drift away from local ones.

The seven crossroads where your network switches

Life comes with a handful of moments when your network reorganises. Each one changes your information flow and your options. Treat them as decisions about networks, not just about places or roles.

Examples: putting network effects to work in your own life

Run a 30-minute network audit

List the 25 to 30 people you spend the most time with, in person or online. Score each column, then look for patterns rather than judging individuals.

A 30-day plan to redesign your network

1. Days 1–3: Audit. Score your 25 to 30 most frequent contacts using the table above, and count how many people sit in each layer. Pick the one layer that matters most right now: the 5 if you feel lonely, the 50 if your career is stuck, the 150 if you want more luck.
2. Days 4–10: Protect your inner circle. Put a recurring slot in the diary with each of your closest five, even if it is a 20-minute call. These hours matter most for health and happiness.
3. Days 11–17: Add one recurring environment. Join a weekly group close to home or work where the people you want to become already gather. Commit to 12 weeks before judging it.
4. Days 18–24: Reactivate weak ties. Message ten moderately weak ties with something useful, and make at least two introductions between people who should know each other.
5. Days 25–30: Create a node. Host one small gathering of four to six people, or start a shared ritual such as a monthly walk or dinner. Then curate your digital inputs: keep one or two high-signal communities and mute what drains you.

FAQ

Is it cynical to think about friends as a network?

Not if the goal is to give more. The research points to shared activity, regular time and helping first, not collecting contacts. Designing your week so you see the people who matter is care, not calculation.

How many close friends do I need?

Dunbar’s work suggests most people sustain around five very close relationships and about 15 close ones. The quality and regularity of those ties matters far more than the total number of people you know.

What is the fastest way to change my opportunities?

Reactivate moderately weak ties and add one new recurring environment. Weak ties bring new information, and a weekly group builds the hours that turn strangers into allies.

Closing thought

Networks are where compounding lives. Trust leads to introductions, introductions lead to opportunities, and opportunities build reputation, which earns more trust. You cannot control every outcome, but you can choose where you spend your hours and who you spend them with. Change one recurring environment this month, and let the network effect do the rest.

Sources: Survey Center on American Life, via ScienceBlog · Holt-Lunstad et al., PLOS Medicine · US Surgeon General, social connection advisory · Chetty et al., Social Capital I · University of Kansas on Hall · MIT IDE on the LinkedIn weak ties study · Fowler and Christakis, BMJ · Centola et al., Science · Rosenfeld et al., PNAS.

The post Network Effects in Real Life: How Your Network Shapes Friends, Careers and Culture first appeared on Activo Group.

Most AI agents fail in production for design reasons, not because the model is weak. The right design depends on the job, the risk and where the data lives. This catalogue covers the single-agent patterns you will actually use, the latest evidence on when several agents beat one, and the orchestration patterns that let agents collaborate without chaos. It ends with worked team designs and a 30-day plan. Updated October 2026.

Why agents fail: four design mistakes

Multi-agent systems add a new class of failure. A 2025 Berkeley study (MAST) analysed over 200 execution traces from seven open-source multi-agent frameworks and grouped the failures into three families.

The building blocks every pattern rewires

Every pattern below is a different way of wiring the same parts: a trigger, a model with instructions, tools, retrieval, memory and an output. Around them sit the control parts that make it safe: evaluators, verifiers, budgets, human checkpoints and tracing.

Single-agent patterns: the catalogue

Start here. Anthropic’s guidance and OpenAI’s guide to building agents agree on one point: get as far as you can with one well-equipped agent before adding more.

When should you use more than one agent? The latest evidence

The debate moved fast between 2025 and 2026. The evidence now points to a clear rule: multiple agents pay off when work can be split into independent parts, and hurt when every step depends on the last.

The standards layer is settling too. MCP connects agents to tools and data, with over 10,000 published servers, and the Agent2Agent protocol (A2A) lets agents from different vendors delegate to each other, backed by over 150 organisations. Both now sit under the Linux Foundation. Microsoft’s Agent Framework names five standard orchestrations: sequential, concurrent, group chat, handoff and magentic.

Multi-agent orchestration patterns

Pick the right orchestration in five questions

Ten design rules for agents that collaborate

Worked examples: multi-agent teams you can copy

Climb the autonomy ladder one rung at a time

A 30-day plan to design your first multi-agent system

1. Days 1–5: Pick one workflow and baseline it. Choose a job with a measurable outcome. Collect 20 real cases and record time, cost and error rate today.
2. Days 6–10: Build the best single agent. Connect the system of record, keep tools few and distinct, add a verifier in front of every write and set budgets. Score it on your 20 cases.
3. Days 11–15: Find the split. Look for subtasks that run independently or need different tools and permissions. If there are none, stop: your single agent is the answer.
4. Days 16–22: Add orchestration. Choose the pattern from the five questions above. Write a brief template for every handoff, assign ownership and set up a shared task list and run ID.
5. Days 23–27: Evaluate end to end. Run the 20 cases through both designs and compare outcome quality, cost and time. Read traces for repeated steps, lost context and weak checks.
6. Days 28–30: Ship at the right rung. Launch as a copilot or supervised team, set weekly review of traces and escalations, and agree the evidence needed to climb a rung.

Closing thought

The best agent systems look less like a swarm and more like a well-run team: clear roles, written briefs, one source of truth, someone who checks the work and a manager who knows when to stop. Start with one strong agent, add collaborators only where the work truly splits, and let evaluation, not enthusiasm, decide how much autonomy each agent earns.

Sources: Anthropic, How we built our multi-agent research system · Anthropic, Building effective agents · Cognition, Don’t build multi-agents · Cemri et al., Why Do Multi-Agent LLM Systems Fail? · Google Research, Towards a science of scaling agent systems · Simon Willison on Cursor’s long-running agents · Addy Osmani on Claude Code agent teams · Linux Foundation, Agentic AI Foundation · Linux Foundation, A2A one year on.

The post AI Agent Design Patterns: The Practical Catalogue (2026) first appeared on Activo Group.

AI startups create more value than any previous software generation, yet many capture only a fraction of it. Inference costs real money on every task, agents replace the seats that software used to charge for, and competitors can copy features in weeks. In that world, how you charge is part of what you build. This guide covers the latest data on AI pricing, how to price agents and multi-agent systems, the models that work, worked examples and a 90-day plan. Updated October 2026.

Why AI breaks the old pricing playbook

The myth of the technical moat

A two-year coding head start used to be a moat. Now AI coding agents can rebuild a feature set in weeks, and the cost of model performance falls by an order of magnitude or more each year. What competitors cannot copy quickly is the commercial structure around the product. Pricing becomes a moat when it does three things:

The 20/80 axiom: do not give away the farm

Often 20% of what you build drives 80% of willingness to pay, and that 20% is usually the core AI capability you shipped first. Give it away to grow and you are left charging for the hard 80% that customers value least. Audit every feature on two numbers: how much it drives willingness to pay, and how much it costs to serve.

Choose a model: autonomy versus attribution

Place your product on two axes. Autonomy is how much work the AI does without a person. Attribution is how clearly you can prove the result it produced. The quadrant points to the pricing model.

A June 2026 survey of 255 B2B software companies by Nalpeiron shows how mixed models have become: 66% use metered usage, 52% sell credit or token bundles, 49% include AI in per-seat plans and 44% use some outcome or value-based pricing. Most companies now run more than one model at once.

Pricing agents and multi-agent systems

Agents change the cost curve. Anthropic reported that a single agent uses about four times the tokens of a chat, and its multi-agent research system about 15 times, because a lead agent runs several subagents in parallel. That made results far better, but it also means the way you orchestrate agents now sets your gross margin.

Recent pricing changes show the market working this out:

Protect margin when your product runs many agents

Worked example, illustrative: a support agent charges $1.00 per resolved ticket and resolves 60% of conversations, so it earns $0.60 per conversation. If one well-tooled agent costs $0.08 per conversation to run, gross margin is about 87%. If an unrouted four-agent team handles every conversation at $0.30, margin falls to 50%. Routing only the hardest 20% of conversations to the team brings the blended cost to about $0.12 and margin back to about 80%.

Pricing ideas that capture more value

Examples: pricing models for your startup

Test willingness to pay before you set the price

Use the Van Westendorp price sensitivity questions in 15 to 20 customer interviews. Ask at what price the product would be so cheap you would doubt its quality, a bargain, getting expensive but still worth considering, and too expensive to consider. The range between bargain and expensive is your acceptable zone. Aim for the upper half when your attribution is strong.

Never run a free pilot. Charge for proofs of concept, even a nominal amount, and frame each one as a joint business case with an agreed success metric and a 90-day deadline. If you cannot show time saved or money made in that window, the problem is product fit, not price.

A 90-day pricing-first plan

1. Days 1–10: Map your quadrant and your costs. Place each product on the autonomy and attribution matrix. Measure the cost to serve per task, including how many agents and tokens each task really uses.
2. Days 11–25: Audit the 20%. Score features on willingness to pay and cost to serve. Move the core AI job behind the paywall and meter anything with highly variable cost.
3. Days 26–40: Test willingness to pay. Run 15 to 20 Van Westendorp interviews and anchor prices to the labour or revenue budget, not the software budget.
4. Days 41–55: Choose your unit and guardrails. Pick seats, credits, outcomes or a hybrid. Write the outcome definition, add spend caps and alerts, and route complex work to multi-agent flows only when needed.
5. Days 56–75: Run paid pilots as business cases. Charge for every proof of concept, agree success metrics upfront and report results weekly.
6. Days 76–90: Launch the attribution dashboard. Show customers the work done and value created, track your own cost per outcome, and review pricing every quarter.

Closing thought

Pricing is not a maths problem to solve after scale. It is part of the product and, increasingly, part of the moat. Charge for the work your AI delivers, design your agent orchestration with cost per outcome in mind, and make the value visible every week. Price for the value you create, or a competitor will.

Sources: SaaStr on ICONIQ State of AI 2026 · Bessemer, State of AI 2025 · Epoch AI, LLM inference price trends · Nalpeiron, 2026 State of B2B Software Monetization · Anthropic, multi-agent research system · Simon Willison on Cursor pricing · eesel on Agentforce pricing · GitHub Docs, Copilot requests · Umbrex on the Van Westendorp method.

The post Mastering AI Monetization: Why Pricing Is a Startup Moat (2026) first appeared on Activo Group.