Diffco

07/30/2026

AI agents vs copilots vs automations: a C-level decision framework for 2026
Vendors stopped distinguishing between agents, copilots, and automations because all three labels sell. So leadership teams get three decks promising the same outcome with three very different risk profiles.

The useful distinction is simply: who is in the loop?

Automation — AI inside a deterministic flow. The model is a component, not the conductor. Copilot — AI in the user's hands during the task. The user has the final keystroke. Agent — AI takes a goal and executes a multi-step plan. The human is at the boundary, not inside the work.

Plot the work on two axes — cost of a wrong answer, and complexity of the path — and the choice makes itself.

High cost of error plus low complexity? Automation with a human review step. Banks, healthcare, legal live almost entirely here.

High cost of error plus high complexity? Copilot. The human in the loop is the product.

The year's most expensive failure mode is building an agent for high-cost-of-error work because agents are fashionable. Agents aren't a higher form of AI. They're a different product shape.

Match the shape to the work.
https://diffco.us/blog/ai-agents-vs-copilots-vs-automations-a-c-level-decision-framework-for-2026/

07/30/2026

Multi-agent system architecture: a teardown of how we built a customer support agent
We shipped a customer support agent handling a five-figure number of tickets a week. 84% resolution in month one, 88% by month three.

We did not start by deciding to build a multi-agent system. We started with a problem and an eval set, and got there over four iterations because single-agent kept hitting walls.

Iteration 1: one model, one 4,000-token prompt, all five ticket categories. Instructions bled across categories. Middling accuracy, premium rates for what should have been cheap classification.

Iteration 2: router plus one solver. Bleed-through stopped, complex categories still failed.

Iteration 3: router plus specialists. 58% → 81% in a week. Then we noticed 7% of tickets touch two categories and the router was throwing half the answer away.

Iteration 4, shipped: triage classifier, parallel specialists, orchestrator that stitches structured outputs, reviewer, and a confirmation step.

Deliberately unfashionable choices — agents communicate in JSON, not English. No autonomous actions on money or identity. Strict tool boundaries per agent.

If you're reaching for multi-agent because it's interesting rather than because single-agent failed you, you're going to overspend.
https://diffco.us/blog/multi-agent-system-architecture-how-we-built-a-customer-support-agent/

07/30/2026

"Should we self-host to save money?" lands in our inbox a few times a quarter. Usually from a CFO staring at a bill.

The honest answer is that most teams who run the numbers properly stay on the API — and the ones who don't run the numbers end up paying more for worse output.

What gets left out of the self-hosted estimate: GPU fleets sized for peak sit idle most of the time, so real utilization lands at 30–60% and effective cost is 1.5–3x the headline rate. Then inference engineering — batching, KV-cache, speculative decoding — which is free in API pricing and 3–5x throughput if you skip it. Then the engineering time, the eval ownership, the reliability scar tissue you haven't earned yet.

Break-even usually arrives at sustained hundreds of millions of tokens a day. Not before.

But cost is sometimes the wrong frame entirely. Data residency, sovereign cloud, latency floors, big LoRA portfolios — there the comparison isn't self-hosted vs API, it's self-hosted vs no product.

The strongest architectures we see are hybrid: frontier API for the hard 20%, small self-hosted model for the easy 80%, cache in front, router deciding.

Pick a posture, not a religion.
https://diffco.us/blog/self-hosted-vs-api-based-llms-the-2026-cost-and-control-tradeoff/

07/30/2026

Every fintech leader wants what everyone else wants from AI — faster ops, better customer experience. They just happen to operate under regulator oversight and criminal liability for some failures.

Here's how we sequence AI programs in regulated finance, and the architectural rules we don't bend.

https://diffco.us/blog/ai-in-fintech-building-secure-agentic-workflows-for-regulated-environments/

07/30/2026

A team that's excellent at using AI to write code can be terrible at building products that run on AI. And vice versa.

They're two different crafts, and blurring them is how companies staff and budget the wrong work.

Wrote up the distinction, plus what each one actually rewards in hiring.

https://diffco.us/blog/ai-native-vs-ai-assist-development-two-different-jobs-often-confused/

07/30/2026

Why you still need developers in 2026 — and why "vibe coding everything" is a trap
A vibe-coded product usually lasts 3–9 months before the foundations crack.

Not because the tools are bad. They're remarkable — a clickable prototype in an afternoon, an internal dashboard before lunch. We use them ourselves.

The trap is treating "great for prototypes" as "great for products."

What breaks is boringly predictable. Schemas that are roughly right and exactly wrong. Auth that looks plausible and isn't. No tests, no observability, and nobody holding the system in their head. Feature velocity halves, then halves again, and someone eventually rebuilds under time pressure with paying customers watching.

Here's the part that surprises founders: one developer brought in during the first 60 days usually prevents the crack entirely. That cost is a fraction of the rebuild.

Developers in 2026 type less code than they did in 2024. They make more decisions — and each decision now flows through AI-amplified output, so it compounds faster in both directions.

So don't stop vibe coding. Just don't vibe code forever.

There's a confident view making the rounds in startup circles right now: with the tools available in 2026 — Lovable, Bolt, v0, Cursor, Claude Code, the rest — anyone can build software. Engineers are a legacy expense. The new way is to describe what you want, accept what the AI…

11/07/2025

Our CEO Vadim Peskov joined Making Big Shifts podcast to talk about building faster with AI. He explains how companies can test early, adapt quickly, and lead change with confidence.

Watch the full episode here:

In this episode, Josh Anderson speaks with Vadim Peskov, CEO of Diffco, about the essential shifts entrepreneurs must make to succeed in the ever-evolving la...