Switch

Agent Library

Agent templates you can point at real work.

Every template is broken down the same way: which foundation model runs it, which tools it is connected to, the instructions it works from, what it hands back, where a human signs off — and what the work took before and after.

Marketing & brand

Campaign & image agent

Produces on-brand campaign visuals and product imagery from your existing brand world.

01 — Foundation model

Image generation model (Nano Banana / Seedream class), guided by a text model

02 — Tools connected

Brand asset libraryDrive / SharePointFigmaAd and social platforms

03 — Instructions

Stay inside the brand world: our palette, typography, lighting and product truth. Never invent packaging details. Deliver every channel format in one pass.

04 — What it gives back

  • Campaign key visuals
  • Format variants for each channel
  • Retouched product shots

05 — Human checkpoint

Brand owner approves the selected visuals before anything is published.

Before / after

Before

3–4 weeks, studio shoot and retouching

After

Same day, in-house

Social & content

Social content agent

Turns a single campaign idea into a full week of channel-specific posts and captions.

01 — Foundation model

General text model with reasoning for angle selection

02 — Tools connected

Tone-of-voice guidePast post archiveSocial schedulerSlack

03 — Instructions

Write in our house voice. One idea, many angles — never repeat a hook. Respect each channel's length and format. Flag any claim that needs legal review.

04 — What it gives back

  • Post copy per channel
  • Hooks and variants for testing
  • Suggested visuals

05 — Human checkpoint

Social lead edits and schedules the final set.

Before / after

Before

A full day of writing per week

After

15 minutes of editing

E-commerce

Product content agent

Writes complete product content in your catalogue structure, ready to import.

01 — Foundation model

General text model, multimodal for reading spec sheets and product images

02 — Tools connected

PIMCommerce platform (Shopify etc.)SEO keyword listSpreadsheets

03 — Instructions

Follow the category template exactly. Only state facts present in the spec sheet. Hit the SEO rules without keyword stuffing. Output in our import columns.

04 — What it gives back

  • Long and short descriptions
  • Attributes and bullet points
  • SEO title and meta text

05 — Human checkpoint

Category manager spot-checks a sample before the bulk import.

Before / after

Before

20 products a day, written by hand

After

200 products an hour

Localisation

Localisation agent

Localises product and campaign content per market instead of translating word for word.

01 — Foundation model

General text model with a long context window

02 — Tools connected

Market glossaryTranslation memoryPIMCMS

03 — Instructions

Localise, don't translate. Use the market glossary for every regulated term. Never soften or strengthen a legal claim. Flag anything that doesn't travel.

04 — What it gives back

  • Per-market content sets
  • Glossary-consistent terminology
  • Flagged legal risk lines

05 — Human checkpoint

Local market lead signs off before publishing.

Before / after

Before

2 weeks per market via an agency

After

1 day in-house

Buying & assortment

Assortment analysis agent

Turns sales and stock data into a decision-ready assortment review.

01 — Foundation model

Reasoning model with data analysis and chart generation

02 — Tools connected

ERPBI exportsExcel / SheetsLast review as a template

03 — Instructions

Compare against category targets, not last year alone. Name the delist and reorder candidates explicitly. Show the number behind every recommendation.

04 — What it gives back

  • Performance summary per category
  • Delist and reorder candidates
  • Slide-ready charts

05 — Human checkpoint

Buyer validates the recommendations before the category meeting.

Before / after

Before

2 weeks of preparation

After

An afternoon

Customer service

Reply & ticketing agent

Drafts an answer for every incoming ticket and routes it to the right queue.

01 — Foundation model

Fast general text model, tuned for tone and consistency

02 — Tools connected

Helpdesk (Zendesk, Freshdesk, Intercom)Help centreOrder systemEmail

03 — Instructions

Answer only from our policies and the customer's order data. Never promise a refund or exception. If confidence is low, escalate instead of guessing.

04 — What it gives back

  • Draft reply in your tone
  • Suggested category and priority
  • Escalation note when unsure

05 — Human checkpoint

The agent reads and sends — nothing goes out unreviewed.

Before / after

Before

8 minutes per ticket

After

90 seconds per ticket

Sales

Proposal & offer agent

Assembles a tailored proposal from meeting notes, pricing rules and past winning decks.

01 — Foundation model

Reasoning model with a long context window

02 — Tools connected

CRMPrice list and termsReference case libraryDrive / SharePoint

03 — Instructions

Mirror the customer's own words on the problem. Use approved pricing only. Never invent a reference or a delivery date. Keep it to two pages.

04 — What it gives back

  • Draft proposal document
  • Scope and pricing section
  • Follow-up email

05 — Human checkpoint

Salesperson adjusts scope and pricing before sending.

Before / after

Before

Half a day per proposal

After

20 minutes

Finance & reporting

Reporting agent

Builds your recurring reports on schedule and writes the commentary around the numbers.

01 — Foundation model

Reasoning model with data analysis

02 — Tools connected

ERP / accounting systemBI toolsExcel / SheetsReport template

03 — Instructions

Never restate a number the source doesn't contain. Comment on variances above the agreed threshold. Follow the template exactly, month after month.

04 — What it gives back

  • Finished report in your format
  • Written variance commentary
  • Highlighted anomalies

05 — Human checkpoint

Controller reviews the numbers and commentary before distribution.

Before / after

Before

3 days of assembly every month

After

30 minutes of review

Meetings & follow-up

Meeting-to-actions agent

Converts a recording or transcript into decisions, owners and deadlines.

01 — Foundation model

Speech-to-text model plus a general text model

02 — Tools connected

Teams / Google MeetTask toolsSlackCalendar

03 — Instructions

Separate decisions from discussion. Every action needs an owner and a date. Quote the transcript when a decision is ambiguous rather than resolving it.

04 — What it gives back

  • Short summary
  • Decision log
  • Action list with owners

05 — Human checkpoint

Meeting owner confirms the actions before they are distributed.

Before / after

Before

45 minutes of note-writing per meeting

After

Instant, from the recording

Legal & admin

Document review agent

Reads long documents and surfaces obligations, risks and deviations from your standard terms.

01 — Foundation model

Reasoning model with a long context window

02 — Tools connected

Contract archiveYour standard termsDrive / SharePointEmail

03 — Instructions

Compare clause by clause against our standard terms. List deviations, never opinions. Quote the exact wording. Flag anything you are unsure about.

04 — What it gives back

  • Plain-language summary
  • Deviation and risk list
  • Suggested questions to the counterparty

05 — Human checkpoint

Legal or the responsible owner makes every final call.

Before / after

Before

A day of reading per contract

After

10 minutes on the flagged points

Research & insight

Market research agent

Compiles competitor, category and market briefs from public sources you can verify.

01 — Foundation model

Reasoning model with web search

02 — Tools connected

Web searchSource allowlistSlackDrive

03 — Instructions

Cite every claim with a link. Prefer primary sources. Say clearly when something could not be verified instead of filling the gap.

04 — What it gives back

  • Structured brief with citations
  • Competitor comparison table
  • Weekly change digest

05 — Human checkpoint

Analyst verifies the cited sources before circulating.

Before / after

Before

A week of desk research

After

An hour, with sources

HR & communications

Internal comms agent

Writes internal communications and onboarding material in your organisation's voice.

01 — Foundation model

General text model

02 — Tools connected

IntranetSlack / TeamsEmailTone-of-voice guide

03 — Instructions

Lead with what changes for the reader. No corporate filler. Same message adapted per channel and per audience, never contradicting itself.

04 — What it gives back

  • Announcement in every channel format
  • Onboarding and FAQ material
  • Manager talking points

05 — Human checkpoint

Comms owner approves before it goes to the organisation.

Before / after

Before

2 days of drafting and review rounds

After

An hour

Templates are starting points, not a shelf product. In the identification workshop we map your actual workflows and propose the agents that create the most value first — then build them around how your team already works.

How it feels to use one

Like messaging a colleague who is never busy.

  1. Step 01

    Give the task

    You tell the agent what you need — like a message to a colleague.

    ~2 min

  2. Step 02

    Result now

    The output appears in front of you. No queue, no waiting.

    30–60 s

  3. Step 03

    Iterate now

    Changes come back in seconds — as many rounds as you want.

    Minutes

  4. Step 04

    Done

    Same quality, your standards — without the hourly invoice.

    ~10 min total

Next step

Which agent should you build first?

Book 30 free minutes with Jakob or Pontus. Bring one workflow that annoys your team and we will tell you whether an agent can take it.

Book a free consultation