Hi everyone,

Still can’t believe it has been about 3 weeks since getting back from the states. In addition to attending Oktane, there have been a few meetings with contacts that I have known and worked with over the years. One will turn into a new collaboration where we are hoping to get our start next month. The other conversations touched on integration trends and the AI impact.

Contrary to popular belief, the integration work still exists. While there was the concern about AI completely replacing it, we are beginning to see two separate trends develop:

  1. Finding out that relying on AI to generate the integration doesn’t work as a complete replacement so teams are now coming back in to “fix” the mess that was created.

  2. Agentic systems are still considered a magical team of coders, but what are they really doing?

Platforms like Claude Code and even Loveable have brought custom software to the masses in a way that has opened up a new world of possibilities. This is what has attracted many people (myself included) to the profession: the magic of seeing something you created work for the first time.

The problem is how easy it is to get seduced into thinking how the current generative AI world is an expert at everything. As Jonathan Stark commented recently on, it is at best a summer intern. They are over-eager with what they produce, but there still needs editing from someone who knows what they’re doing. Also, you cannot expect an intern to be an expert: their default mode is to act unless you explicitly tell the tool to ask clarifying questions. Even then, the quality of the questions doesn’t get to the true nature of what is needed.

And this is where you need a team to come in to address the gaps.

The second trend crosses an observation I made initially at the conference where, whether it is on a task basis, or a per-agent/tool basis, we will need to find a good way to handle the user/permissions model. Not just for the company day to day operations, but also thinking about the acquisition of two companies leveraging AI agents, how each side manages them from a governance standpoint, and aligning their functionality to reduce duplication in functionality.

We are still in an uncharted world from this standpoint along with far too much hype to even have have a common dictionary to use for us to have an intelligent conversation about the problem. This will become even more critical depending on whether we treat the agent as a one-off task to purchase plane tickets for an executive. Perhaps as an independent task watching the general ledger and flagging transactions for wasteful spending. Or it could be a group of agents responsible for pulling security updates from a known database and constantly making updates, testing, and pushing it out to production.

As software is just codified processes within an organization, and each organization has their own process, you can see how this can get pretty complex fast. Especially when operating on an expectantly reduced timeline.

This second trend is the one that I do see growing over the next year. Especially as the hype cycle begins to die down, and patterns begin to develop on how best to approach this new world.

Partially because of this, I’ve been working on creating a podcast that is focused more on the M&A integration front. We are planning to launch at the end of next month, and I hope to have some more details to share in the meantime.

cab