Industries / Technology / Software and SaaS
ERP and CRM software
ERP and CRM software is the packaged software that companies use to run their finance, purchasing, stock and production from one shared set of records (enterprise resource planning, ERP) and to keep track of their customers, sales and service requests (customer relationship management, CRM).
1972
On 1 April 1972 five former IBM employees, Dietmar Hopp, Hasso Plattner, Claus Wellenreuther, Klaus Tschira and Hans-Werner Hector, started the company now called SAP in Germany, to write standard software that joined up a company's business processes and processed its data in real time.
34
Milestones achieved across 6 domains.
13
To watch for: announced, not yet confirmed outside.
12
Open deliverables. Nobody has shipped them.
In one minute
Where it started
In 1972 five former IBM employees founded SAP in Germany to sell the same accounting and purchasing software to many companies, and to process each entry as it happened instead of in overnight batches. Factory planners were already using computers to work out which parts to order and when, a method called material requirements planning (MRP), and by the early 1990s those planning systems had grown into enterprise resource planning, which added accounting, sales and staff records. Customer relationship management began in the late 1980s with programs that kept one salesperson's contacts, and in 1995 Siebel Systems put a whole company's customers and deals into one shared database.
Where it stands
SAP reported revenue of €36.8 billion for 2025, €21.0 billion of it from cloud subscriptions, and Salesforce reported $41.5 billion for the year to January 2026. Much of today's implementation work is moving companies off SAP's older ERP, whose mainstream maintenance ends in 2027; by September 2024 Gartner counted only 37% of its customers as having licensed the successor, according to The Register. SAP, Oracle, Microsoft and Salesforce all now sell AI agents inside their software, while public tests find the best agents completing about 58% of CRM tasks given in one message and 35% when the task unfolds over a conversation.
What is open
The costliest failures, from the US Air Force's canceled logistics system to Birmingham City Council's Oracle rollout, came from changing standard software to fit old ways of working, and the vendors' new agents come with figures the vendors measured themselves. The open jobs below are public tests that would check such work: invoices matched against orders, conflicting duties in user roles, supplier emails turned into order changes, migrated data checked for planted errors, community modules ported by agents, and agents tested against approval limits. Most can be started this month with open-source ERPs such as ERPNext and Odoo, made-up company data and current AI models, on systems the builder runs.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Finance
The ledger and the work that feeds it: recording sales and purchases, checking and paying supplier invoices, collecting from customers, and keeping the controls that stop errors and fraud.
5 · 2 · 2
Last achievedOld flaws in live systems2016
Watch forInvoices matched by an agent
2
openSupply chain
Deciding what to buy and make, and when: forecasting demand, planning materials and factory capacity, placing purchase orders with suppliers and keeping the stock records the plan depends on.
6 · 2 · 2
Last achievedAgents tested on factory work2026
Watch forA simulated year, no stockouts
2
openSales
Sales software keeps the record of a company's customers, their contacts and every deal in progress, so that sellers and their managers can see who to call next and what is likely to close.
5 · 2 · 2
Last achievedPhone calls empty the CRM2025
Watch forAn agent for every lead
2
openCustomer service
Service software keeps the record of each customer's problem, called a case or a ticket, from the first message to the fix, along with the answers that support staff, and now AI agents, reuse.
5 · 2 · 2
Last achievedPaying per solved case2024
Watch forMost requests handled by AI
2
openPutting it in
Putting it in is the work of choosing a new ERP or CRM system, setting it up, loading the old system's data into it and switching it on, usually with outside consultants, and then keeping it on a version the vendor still supports.
7 · 3 · 2
Last achievedAn agent ports ERP code2026
Watch forA third less migration work
2
openAI agents
AI agents are programs built on large language models that read and change the records in ERP and CRM systems on a person's behalf, for example by answering a customer, qualifying a sales lead or entering an order.
6 · 2 · 2
Last achievedResults checked in the database2026
Watch forOrder exceptions in minutes
2
open
Advances to build with
4 research notes

Muse Ember
The jobs agents repeat
Meta's Muse Ember matched Claude Haiku 5.5 on an outside firm's tests at a third of the cost per task, though both list the same prices for text in and out.

GitHub Copilot Fusion
Effort, not model
GitHub Copilot Fusion picks the AI models for each request, and in a test GitHub paid for, its top setting scored 2 points below a rival at about half the cost.

Cursor Agent SDK
Findings that cite notes
Cursor released a kit for building AI agents, and its example, a code reviewer that cites the note behind each finding, is known only from Cursor's account.

Microsoft ThinkingBox
All twenty times
By Microsoft's own count, 50 steps of training took an AI model from passing 38 of 507 test tasks on all 20 tries to passing 107.
Open in erp and crm software
3 of 12 open deliverables
Open · Finance
A public three-way match test
A public set of made-up purchase orders, goods receipts and supplier invoices, loaded into an open-source ERP, in which some invoices carry planted problems: a wrong price, a short delivery, a duplicate, the wrong unit of measure, goods never received. It scores how often the ERP's own matching rules, and AI agents, pass the good invoices and stop the bad ones. Matching an invoice to its order and its receipt is called three-way matching.
Built with what exists
ERPNext or Odoo Community, open-source ERPs that hold purchase orders, receipts and supplier invoices; Python with the Faker library to make up suppliers and orders, and a PDF library such as ReportLab to print the invoices, so the data behind each PDF is the answer key; AI agents built on models such as Claude, Gemini and GPT, each run several times (see the Note "All twenty times"), with the cost per invoice recorded (see the Note "The jobs agents repeat").
Open · Supply chain
Supplier emails into orders
A public test set of supplier emails about purchase orders, such as confirmations, new delivery dates, short shipments, price changes, cancellations, split deliveries and replies that promise an answer later, each labeled with the order, line and field it changes. It scores how often AI agents sort each email correctly and propose the right change to the order.
Built with what exists
made-up purchase orders in ERPNext or Odoo Community; emails written from templates and by AI models, then checked by someone who has worked in purchasing; a model that returns probabilities for fixed-answer questions, such as Gemini Decision 1, to sort the emails by type (see the Note "A thousand questions, one pass"); agents built on models such as Claude or GPT to propose the field changes.
Open · Sales
Call notes against a key
A public test of the AI note-takers that listen to sales calls and fill in CRM fields. Record 100 scripted discovery calls, the first real conversation between a seller and a possible buyer, each with a written answer key for eight fields: budget, decision-maker, need, timeline, competitor named, next step, its date and deal size. Play every call through each note-taker and score each field it writes as right, wrong, missing or invented.
Built with what exists
two volunteers reading scripts, or synthetic voices; trial or free accounts of CRMs that write call notes into records; an open pipeline for comparison, made of a speech-to-text model such as Whisper or ElevenLabs Scribe (see the Note "Fewer word errors") and a language model; a spreadsheet for the key and scores.
Where it sits
Broader