Year 4 of AI: Thoughts on the Journey So Far

I was skiing last week and had some AI related thoughts floating around in my head. So this week, let me share them.

First off, some reflections as I was on the slopes.

In November 2022, ChatGPT changed the world. It’s easy to forget that before then, “normal”, non-tech people had no real clue what AI was. Now, kids and adults alike use it daily. It’s ubiquitous. AI was all around me, people using AI search summaries to find a good place to have lunch halfway up the slopes, in my case where to recycle an old ski jacket etc.

And last week, as I was scanning my feeds on a particularly long 20-min ski lift, I came upon OpenClaw (originally named ClawdBot). This open-source project is only two months old, hard to believe. It allows anyone to run AI agents against completely local AI models. The premise is profound: we are entering a world where an AI can handle 90-95% of a given task, completely securely, without a single byte of data leaving your controlled environment. Released exactly 3 years after ChatGPT, it’s amazing how far we’ve come.

So, as we enter Year 4 of what I think is at least a 10-year AI journey, it’s getting real.

1. The New Form Factor: Local, Not Just Cloud

For Capture and Proposal teams operating in regulated industries, the public cloud has always been problematic. Using public AI models are a no-no due to security concerns. So, you must use secure-AI.

OpenClaw for me represents a shift in form factor. You move from "Chatting with a Bot" to "Deploying an Agent", and it’s completely secure.

In the GovCon context, this means the ability to run high-reasoning agents inside your own secure perimeter (or even on a classified edge). It’s the shift from renting intelligence to owning a private, secure digital workforce.

I wonder whether this will encourage IT teams to take open source, fully secure, highly performant AI models more seriously. I suspect that CFOs will certainly be interested, given the economic advantages. Open-source models like Kimi K2.5 will likely see renewed impetus (despite the country of origin).

2. Systems of Record are Going Nowhere

Here’s something else that’s been buzzing in my brain. The financial markets are punishing SaaS vendors right now. There’s a suggestion that they are not needed (possibly to be replaced by vibe-coded mini-apps). Here’s the thing, System of Record products are still absolutely needed in the enterprise.

In the rush to adopt AI, there’s a temptation to think we can bypass "boring" legacy systems. But AI agents are only as good as the data they can access.

Your CRM, your past performance library, and your contract management systems are not going anywhere. They are the Systems of Record. AI is the System of Intelligence that sits on top of them. The winners in the next three years won’t be the firms that replace their databases, but the ones that use AI & agents to make that data useful for their Proposal Managers in real-time.

The key is that they must be open such that AI layers can harness the data contained within. So, expect more hybrid systems, combining systems of record and AI.

3. Eating Your Own Dogfood

We often talk about "transformative change" for our government clients, but are we applying it in our own shops?

Two weeks ago, I used VisibleThread end-to-end to create an RFI (Request for Information) response. It was a massive help and sped up my ability to create high quality answers. I completed the RFI in about a 1/3rd of the time it would have taken manually.

Direct use of VT also pinpointed a few areas of our product where I think we can do better in terms of design & usability. As CPO (Chief Product Officer) that’s a key part of my remit.

And along those lines, from Nov thru Dec, we ran an Early Adopter program for our new platform, VisibleThread 7 with over 30 customers participating. The feedback was incredibly useful there also, and we’ve made many tweaks/adjustments as a result. In short, all vendors should use their own products, so they feel what it’s like to be a user and provide the best possible user experience.

Similarly, if your leadership team isn't "eating their own dogfood", actually using AI agents or AI tools to draft executive summaries or analyze RFPs, they won't understand the friction points your team faces. Better design in AI isn't about prettier buttons; it’s about understanding the "Human Factor" of how a Capture Manager actually interacts with a machine to make a "Go/No-Go" decision.

We should all "eat our own dogfood".

In Summary:

1. The progress AI has made since Nov 2022, and over the last 3+ years has been profound, culminating in OpenClaw as an impressive example of private AI.
2. Systems of record products are 100% needed, they’re going nowhere, but they do need to act in concert with AI systems of intelligence.
3. Using your own products or solutions (aka eating your own dogfood) is a winning strategy.

Till next time.

Best,

Fergal

Founder & CPO (Chief Product Officer) at VisibleThread

---

PS: Please subscribe to The Human Factor biweekly newsletter where I share more insights like this.