When I say “agentic,” I mean the ability to act. An agentic enterprise is one where software agents carry out decision-making and execution on behalf of the company. Think of it as moving from spreadsheets and dashboards, where humans interpret data, to a workplace where AI agents hold the pen, make the call, and initiate the action.

From digitised to agentic

Most companies today call themselves “digitally transformed.” In reality, they are digitised: their processes live in software, but in rigid, predefined flows. For example, when a customer email arrives in a traditional support system, it’s sorted into a queue or tagged as spam. An agentic system, by contrast, reads the email, interprets its intent, and – without waiting for human hands – replies with the right information or escalates intelligently.

I’ve lived this contrast. At ParkStreet, where I served as CTO, we integrated our ERP and payments workflows with an agent:// protocol we published to the IETF. That meant our invoicing wasn’t just automated – it was adaptive. When regulations shifted in the EU, our agents rewired themselves to stay compliant with PEPPOL standards without weeks of developer rewrites.

Technology meets culture

The leap to agentic organisations is not just about sprinkling large language models into existing workflows. It’s a tectonic shift. On the technical side, it demands more than APIs: it requires multi-agent orchestration, semantic graphs, and protocols like MCP (Model Context Protocol) that allow agents to speak with existing software. AI pushes us further, and Continuous Integration becomes Continuous Alignment (CI/CA/CD) – every new prompt, model, or dataset is tested, evaluated, and governed.

But culture matters just as much. I often remind teams: “An AI doesn’t replace you – it reflects you.” Leaders must stop fearing agents and instead treat them as teammates. Employees are no longer task executors; they are reviewers, orchestrators, intent-setters. When I introduced agent-assisted coding across two continents, I told my teams: “You’re not losing control, you’re gaining altitude.” And it worked – the teams went from firefighting sprints to delivering at elite DORA metrics.

Real-world cases

So where are agentic organisations already visible?

• ERP & Payments: at ParkStreet, our monthly payment collections – a task that once consumed weeks – shrank to hours after we connected LLM agents to ERP systems via MCP. The agents not only issued reminders but also negotiated payment plans within policy guardrails
• Healthcare Compliance: at Zedsen, we used AI-assisted simulations for hardware design. Where traditional iterations took years, agents reduced cycles to months, accelerating a cancer diagnostics platform through MDR compliance without cutting corners
• E-commerce and personalisation: at Jio, we didn’t use AI for simple product recommendation; AI helped personalise promotions, discounts, and even coupon strategies at scale, turning static catalogues into adaptive, customer-first experiences

These are glimpses of a future where agents don’t just follow instructions – they collaborate, optimise, and preempt.

The risks we must tame

Of course, this transformation isn’t without peril. I’ve seen AI agents misfire spectacularly – hallucinating reports, misclassifying customers, or worse, deleting critical datasets when permissions weren’t locked down. That’s why guardrails matter. Observability, traceability, and human oversight have to stay at the centre. Agents may be fast, but they are also stochastic. Garbage data in still means garbage decisions out.

But the hardest challenge isn’t technical – it’s human. Legacy systems can be patched; culture cannot. If leadership treats agents as enemies, adoption stalls. If employees aren’t reskilled as orchestrators, workflows collapse. As I tell CEOs: “Agent AI lifts the floor, not the ceiling. That ceiling is still human ingenuity.”

Why startups have the edge

In all this, startups are the natural pioneers. When I mentor Antler-backed founders, I see their advantage clearly: no legacy baggage, no entrenched workflows, no cultural debt. They are already “vibe coding”: letting LLMs draft sales pitches, chain meeting summaries, and build presentations overnight. For them, agentic isn’t a future state, it’s Tuesday afternoon. That’s why I believe the first “neotic unicorn” – a billion-dollar business run by fewer than ten people with agents as core workforce – will emerge before 2027.

Practical first steps

For established enterprises, here is my simple advice: aim small, miss small. It's better to start with high-value, low-risk workflows. Connect agents to expense verification or resume reviews before you attempt customer-facing systems, and build an interoperability foundation: MCP servers, agentic gateways and semantic layers. Establish guardrails: escalation thresholds, red team evaluations, and human-in-the-loop checkpoints. And most importantly, measure not AI vanity metrics, but business outcomes: cycle time, error reduction, human hours freed.

The road ahead

In the next three to five years, I see most organisations moving into “semi-agentic” or “demi-agentic” maturity levels. Laws will evolve, compliance regimes will adapt, and we may even see non-human directors in some jurisdictions. The greatest barrier isn’t model performance – it’s human adaptability. But for those who embrace it, agentic organisations offer something no ERP, CRM, or BI dashboard ever could: a workplace where intelligence flows at the speed of light, and decision-making scales as fast as compute.

Digital transformation was about making information accessible. Agentic transformation is about making intelligence actionable. The companies that realise this will not just be faster – they will be freer. And in business, as in life, freedom is the ultimate competitive edge.

For more startup news, check out the other articles on the website, and subscribe to the magazine for free. Listen to The Cereal Entrepreneur podcast for more interviews with entrepreneurs and big-hitters in the startup ecosystem.