Services an AI can write and a team can run

75%

of Google's new code is AI-generated

Up from 25% in 2024. Sundar Pichai, April 2026.

84%

of developers use or plan to use AI tools

Stack Overflow Developer Survey 2025, 49,000+ respondents.

53%

of Go developers use AI tools every day

Go Developer Survey 2025, published January 2026.

56%

of AI-written code passes security tests

No better than a year ago. It compiles almost every time.

5x

longer median review time

Pull requests are 51% larger on high-AI teams.

3x

more incidents per pull request

Against 34% more tasks delivered.

46%

of developers distrust AI accuracy

33% trust it. 3% trust it highly.

func main() {

    app := gofr.New()

    app.GET("/orders/{id}", getOrder)

    app.Run()

}

 

func getOrder(c *gofr.Context) (any, error) {

    var status string

    err := c.SQL.QueryRowContext(c,

        "SELECT status FROM orders WHERE id=?",

        c.PathParam("id")).Scan(&status)

    return status, err

}

Running, and nobody wrote it

Traces. OpenTelemetry spans for the request and the query; one env var ships them.

Metrics. Prometheus at /metrics on port 2121.

Logs. JSON lines carrying the trace_id.

Health. /.well-known/health and /.well-known/alive.

Profiling. pprof on the metrics port.

Shutdown. Graceful on SIGTERM, 30 seconds by default.

app.AddLLM(&llm.Client{

    Provider: llm.Groq,

    Model:    "llama-3.3-70b-versatile",

})

 

func ask(c *gofr.Context) (any, error) {

    return c.LLM().Generate(c, prompt)

}

Agent

Any MCP client

app.EnableMCP()

MCP server on 127.0.0.1:8200

Your handlers

Same router, same middleware

Read-only

Only GET, HEAD, OPTIONS and QUERY routes become tools. Writes are never exposed.

Local by default

It listens on loopback, on its own port, and you can exclude routes.

Same rules

Tool calls run through the real router, with the caller's headers forwarded.

1

curl localhost:8000/inventory/A1

A trace and a metric appear. Nobody asked for them.

2

curl -X POST localhost:8000/ask -d '{"prompt":"..."}'

The LLM call is a span in the same trace, tokens counted.

3

MCP client -> localhost:8200

The agent finds /inventory as a tool and calls it.

Serve

HTTP, gRPC, GraphQL, WebSocket, cron jobs, CLI apps, CRUD from a struct

Data

MySQL, Postgres, SQLite, Redis, Mongo, Cassandra, ClickHouse and more, with migrations

Messaging

Kafka, NATS, Google Pub/Sub, MQTT, SQS, Event Hubs, Redis

Files

Local disk, S3, GCS, Azure, FTP and SFTP behind one file API

Observe

Traces, metrics, JSON logs, health endpoints, pprof, live log-level change

Secure

Basic auth, API keys, OAuth with JWT, RBAC, rate limiting, CORS

Call out

HTTP client with circuit breaker, retry, rate limit and health checks

AI

LLM datasource, streaming, embeddings, tool calls, MCP server