AI moves fast. Most organisations don't.
AI capability is growing exponentially. Most companies are still working out what to do with last year's tools. That gap is where the next few years of competitive advantage will be won or lost.
Jacob Bergdahl works hands-on with Monterro's portfolio companies on AI adoption. In this post, he sums up where agentic AI stands right now and what it means for how software companies work.
AI now works for hours, not seconds
The length of task a large language model can complete on its own has doubled roughly every six months. Lately it doubles every few months. Models are arriving that can work on complex problems for days or even weeks.
Today's AI can take a task, break it into parts and hand those parts to several agents that work together until the job is done. That might be a 25-minute presentation or a feature that takes two days to build. Developers got there first. This year, the same capability reached everyone else.
From chat to AI colleagues
If there is one concept to understand this year, it is skills. A skill is a reusable set of instructions that captures a workflow, complete with templates, logos and reference material. Unlike older custom GPTs or Copilot agents, a skill triggers on its own when the context calls for it. Mention a slide deck and the deck skill kicks in.
We see companies move through the same stages:
Ad hoc conversations. People ask their AI for help, one task at a time.
Personal skills. People turn good results into skills. Then they make 10, 50, 500 of them, and nobody knows which ones exist.
Shared ways of working. Teams consolidate. The sales playbook becomes a skill, so the third-best salesperson works like the best one. Skills are grouped into plugins, and the company sets up a shared knowledge base.
Scheduled work. AI runs overnight and over weekends. On Monday morning, follow-ups, call prep and account research are waiting as drafts.
Autonomous AI colleagues. Agents triggered by events, such as a new lead, with their own system permissions. They read and post in Teams or Slack and work around the clock.
Many of Monterro's portfolio companies are moving toward that last stage. Typical jobs for these AI colleagues include scoring incoming leads, running pipeline reviews throughout the day and handling meeting follow-ups.
Hard-won hint: It's easy to automate inefficiencies. Layering AI on top of today's workflow is the simple path. Rethinking the workflow from scratch is harder, and that's where the real gains are.
Everyone becomes a manager
Work with agents for a while and you notice a shift. You delegate. You review. You approve. People who have never managed anyone now run a small team of AI agents.
Three things follow from that:
Generalists win. Workflows used to pass between specialists in long chains of handovers. AI covers so many areas that people with broad experience across a workflow can now run more of it themselves.
Task switching is a core skill. You send one agent off for 15 minutes, start another on a 25-minute job, approve a request from the first and review the second. It's tiring at first. The interfaces for managing dozens of agents at once barely exist yet.
Humans stay accountable. If an agent makes a mistake, the person who deployed it owns it. Quality control can't slip because AI did the first draft. The machine knows the way. Often, but not always.
AI is becoming the user of your software
All the mainstream AI tools can now connect directly to the apps people use every day: Microsoft 365, Figma, HubSpot, Salesforce and many more. That changes how software gets used.
A designer can ask an agent to review a Figma file without opening Figma. A salesperson can let an agent fetch and update CRM records and stop logging in to the CRM altogether. In more and more cases, the AI is the one clicking around inside the system. People think less about which app to open and more about which outcome they need.
The web is shifting too. According to Cloudflare, AI agent traffic overtook human traffic on the internet this summer. Websites now need files like llms.txt so agents can find their way around, and e-commerce needs to work for agents that buy on someone's behalf.
For B2B software companies, this means one thing above all: build MCP servers. Model Context Protocol is the standard that lets AI tools connect to a system. Digitally mature customers will expect it.
Where to start
Look for the sweet spot. AI creates the most value where three things overlap: the work is economically valuable, it's reasonably complex and the cost of an error is low. Software engineering is the obvious example. Code is expensive to write, and a human catches mistakes in review. Plenty of similar sweet spots sit undiscovered in other functions.
Build for change. If you design your organisation around today's tools, it will be outdated by the time it's in place. Keep skills and connectors outside any single vendor, for example in GitHub. Skills are an open standard, so the same skill works in Claude, ChatGPT and Copilot. Avoid lock-in.
Appoint AI champions. Find one person per team who is curious about AI. Give them time and budget to learn, and let them help their colleagues. Have the champions meet every few weeks to share what works. Larger companies can go further and hire someone whose whole job is to help every team get going. At Monterro, that person spent his first six months sitting with finance, sales and everyone else, building skills and connecting systems.
Take data seriously. GDPR compliance varies between tools, and each company needs a clear policy on what data goes where.
AI won't replace people. People using AI will.
So far, the companies we work with that use AI set bigger goals and enter new markets faster. The roles change, though. Routine tasks disappear, and the share of judgement calls goes up.
The rules are being rewritten right now. Most organisations haven't noticed yet. The ones that start using agentic AI across the whole company today, with shared skills, scheduled work and clear quality control, will be hard to catch.
By Jacob Bergdahl, AI Specialist at Monterro
