Forum3 AI-First Newsletter

It was just three weeks ago, we were breaking down the battle of the LLMs over Super Bowl ads. GPT vs. Claude vs. Gemini. Bigger models. Better benchmarks. Faster releases.

But that game is changing, and it’s no longer just about who has the smartest model.

It’s about:

Calendars. Inboxes. CRMs. Docs. Approvals. Audit logs. The entire enterprise stack.

That’s the shift.

We’ve officially moved from the battle of the models… to the battle of the agent platforms.

And this week, the agent platforms started putting real points on the board.

This isn’t another model upgrade cycle. It’s a reset of the playing field. Every major player just revealed their blueprint for the agent era and it’s all about control, orchestration, governance, and integration.

On February 5th, OpenAI launched Frontier — a full enterprise platform for building, deploying, and managing agents with governance and control.

Early customers include Uber, State Farm, Intuit, Thermo Fisher.

The bold move? Frontier is model-agnostic. It supports agents built on Google, Microsoft — even Anthropic.

The rollout of Frontier has already rattled traditional SaaS players. If an agent can execute a sales workflow without a human logging into Salesforce…what happens to per-seat licensing?

Fortune called Frontier a bid to become the “operating system of the enterprise.”

Read more about Frontier here:

For now, Anthropic is taking a different angle and focusing on daily utility. Claude Cowork has moved from impressive demo to real recurring work:

Weekly reports. Recurring briefs.
File organization. Multi-step workflows. Memory portability.

Underneath it all is something bigger: MCP (Model Context Protocol).

MCP is becoming the plumbing of the agent economy. It’s a universal way for agents to access Slack, GitHub, Postgres, Drive, and more without custom integrations.

It’s boring. It’s foundational. But it’s incredibly important.

And the capability leap is real: Claude Sonnet 4.6 hit 72.5% on the OSWorld computer-use benchmark — up from 22% a year ago.

Meanwhile, Claude Code has just crossed $2.5B ARR and now accounts for over half of Anthropic’s enterprise revenue.

Some analysts are calling this a “Software Apocalypse.” Dramatic? Maybe. But the direction is clear.

Here’s an 8 minute video that walks thru Claude’s recent update:

Perplexity’s “Computer” announcement might be the most forward-looking move of the bunch.

It coordinates 19 different AI models into one cohesive system.

Claude for reasoning.
Gemini for deep research.
GPT for long context.
Specialized models for image and video.

The user describes the outcome and the system handles the “swarm”. A swarm of agents is a coordinated group of specialized AI agents that work together to complete a larger objective — dividing the work, sharing context, and handing tasks off to one another automatically.

CEO Aravind Srinivas said:

This isn’t “one model to rule them all”, but more like using the right brain for the right job and at the right time.

Multi-model is normal now. Orchestration is where the value lives.

Read more about Perplexity Computer here:

Then there’s OpenClaw. The fastest-growing open-source AI project ever.

145,000 GitHub stars in weeks.
Mac mini shortages.
A rebrand war.
An OpenAI acqui-hire.

OpenClaw represents something bigger than hype:

It’s the shift from chatbot → agent runtime.

Users aren’t just chatting with it. They’re running:

One user wrote:

And OpenAI’s move to bring the founder in, while committing to keep it model-agnostic under a foundation, signals something bigger:

The future is multi-agent. And open-source will be part of it.

As agents move from suggesting to doing, the risk profile changes.

This isn’t just text generation anymore. It’s execution.

The right approach is simple:

Start in a sandbox (if you can).
Limit permissions. Isolate the environment.

Always supervise.
Begin read-only. Then suggested actions. Then supervised execution.

Experiment small first.
Draft the report. Summarize the inbox. Flag anomalies.
Earn trust before expanding access.

In the agent era, success won’t come from recklessness or hesitation, it will come from teams that move quickly and thoughtfully. Learn fast. Experiment deliberately. Expand control intentionally. Scale with confidence.

It’s clear this is no longer a battle of the model, but a systems race.

Integration. Control. Orchestration. Governance.

The models got us on the field, but the agent platforms will decide what happens next.

And if the last few weeks are any indication, the pace isn’t slowing down.

So buckle up.

We’re barely a few minutes into the first quarter.

Onward.