How we got here 2024-2026
Twelve retrospectives on the years that took AI from chat to agents. All were written in 2026, looking back; each one links its sources.
12 retrospectives
2024 · Data and people
2024: RAG goes mainstream
Bytes4AI · Published · Written in 2026 · 2 min read
Retrieval-augmented generation became the default way enterprises put their own data in front of a model, and retrieval quality became the real work.
2024 · Engineering
2024: Open-weight models close the gap
Bytes4AI · Published · Written in 2026 · 1 min read
In one year the distance between the best open-weight and the best closed models shrank to a few points, and model choice became an engineering decision.
2024 · Governance
2024: The EU AI Act enters into force
Bytes4AI · Published · Written in 2026 · 1 min read
On 1 August 2024 the first broad AI law took effect, with a staged timetable that turned AI governance from a policy topic into a project plan.
2024 · Observability
2024: Reasoning models change how we evaluate AI
Bytes4AI · Published · Written in 2026 · 1 min read
Models that think before they answer jumped on hard maths and puzzle benchmarks, and made the cost and method of each test part of the result.
2024 · Agents
2024: MCP arrives
Bytes4AI · Published · Written in 2026 · 1 min read
The Model Context Protocol gave AI applications one standard way to reach tools and data, and quietly widened what an agent can touch.
2025 · Agents
2025: The agent hype year
Bytes4AI · Published · Written in 2026 · 1 min read
2025 was sold as the year of AI agents. Many pilots stalled, analysts predicted mass cancellations, and the useful question became which agents earn their keep.
2025 · Engineering
2025: Vibe coding and the quality question
Bytes4AI · Published · Written in 2026 · 1 min read
Writing software by describing it to a model went from a joke to a job description, and the question moved from how fast to how good.
2025 · Governance
2025: Prompt injection and agent security become real
Bytes4AI · Published · Written in 2026 · 1 min read
Prompt injection moved from a research curiosity to real vulnerabilities in shipping products once agents could read untrusted content and act on it.
2025 · Governance
2025: AI regulation obligations start to bite
Bytes4AI · Published · Written in 2026 · 1 min read
The first binding AI obligations applied in 2025, and companies learned that compliance is mostly about evidence they can produce on request.
2026 · Data and people
2026: Shadow AI, agents with your credentials
Bytes4AI · Published · Written in 2026 · 1 min read
Shadow IT became shadow AI: staff brought their own tools, then their own agents, and those agents act with the person's own access.
2026 · Observability
2026: The governance-containment gap
Bytes4AI · Published · Written in 2026 · 1 min read
Most organizations can now watch their AI agents. Far fewer can stop one quickly. That gap is the defining AI risk of 2026.
2026 · Governance
2026: What's next, governed, evidence-backed autonomy
Bytes4AI · Published · Written in 2026 · 2 min read
The next step is not more autonomous agents. It is agents whose every change is reviewed, approved and recorded, so autonomy can grow with the evidence.